Impending Collision Judgment from an Egocentric Perspective in Real and Virtual Environments: A Review

Abstract

The human egocentric perception of approaching objects and the related perceptual processes have been of interest to researchers for several decades. This article gives a literature review on numerous studies that investigated the phenomenon when an object approaches an observer (or the other way around) with the goal to single out factors that influence the perceptual process. A taxonomy of metrics is followed by a breakdown of different experimental measurement methods. Thereinafter, potential factors affecting the judgment of approaching objects are compiled and debated while divided into human factors (e.g., gender, age, and driving experience), compositional factors (e.g., approaching velocity, spatial distance, and observation time), and technical factors (e.g., field of view, stereoscopy, and display contrast). Experimental findings are collated, juxtaposed, and critically discussed. With virtual-reality devices having taken a tremendous developmental leap forward in the past few years, they have been able to gain ground in experimental research. Therefore, special attention in this article is also given to the perception of approaching objects in virtual environments and put in contrast to the perception in reality.

Keywords

egocentric collision perception velocity judgment speed perception time-to-contact time-to-arrival time-to-collision looming tau virtual-reality immersive virtual environment simulator validation pedestrian simulator pedestrian crossing head-mounted display

Introduction

The automotive industry faces new challenges in the course of the vehicles’ automation process. The behavior of all surrounding traffic participants needs to be understood and implemented in the vehicles’ artificial intelligence. This also involves, for example, human perception and judgment of an approaching object, such as an oncoming automotive vehicle. While some researchers explored the neural basis on how an approaching object is processed by humans (Billington, Wilkie, Field, & Wann, 2010; Coull, Vidal, Goulon, Nazarian, & Craig, 2008; Field & Wann, 2005), factors and heuristics that may have an impact on how humans judge an oncoming object are also of profound interest to investigators. While these factors may be related to human individuality or the composition of the environment, the technological components of a virtual setup used for the assessment may also affect the judgment and corrupt the findings to a certain degree. Such virtual environments may be found, for example, in driving simulators or similar virtual-reality (VR) systems. Conquering the entertainment industry over the past few years, VR became an intensely growing field, with recent technological advances opening up new possibilities also for the research community. Investigators analyzing human behavior tap into utilization of VR setups due to their advantages in comparison with real-life experiments regarding participants’ safety, cost-efficiency, and reproducibility of investigated scenarios. However, VR setups initially require an unfortunately often omitted validation process analyzing the degree of similarity between participants’ perceptions and behavior within the virtual environment compared with scenarios in reality. Only then can conclusions be drawn based on data collected in a virtual environment, allowing a transfer of those findings to real environments.

The investigation of factors that influence impending collision judgments experienced its heyday back in the 1980s and 1990s after Lee (1976) had introduced his concept of tau. Nevertheless, this field is gaining much relevance again, for example, in the course of training artificial intelligence in vehicles. The article presented here strives to summarize past findings and to connect them to each other. Furthermore, the impact of virtual environments on the outcome of collision judgments is looked at and discussed critically.

Taxonomy of Metrics

The research community employs a variety of terms when investigating the remaining time between two objects approaching one another. Widespread terms are time-to-contact (e.g., Tresilian, 1991), time-to-passage (e.g., Kaiser & Mowafy, 1993), time-to-arrival (e.g., Schiff & Oldak, 1990), time-to-collision (e.g., Cavallo & Laurent, 1988), and time-to-coincidence (e.g., Groeger & Cavallo, 1991). The subtle distinction between these terms may lay in details:

whether the observer or the object or both of them are moving,

whether the objects are on a collision course or passing each other, and

whether the observer is perceiving the collision from an egocentric perspective or externally and uninvolved.

Hancock and Manser (1998) strived to organize the terminology and to create a formal taxonomy: An overview of different terms and their characteristics based on Hancock's and Manser's suggestion is given in Table 1.

Table 1.

Terminology and Characteristics of Metrics Used to Describe the Remaining Time Between Two Objects on an Approaching Course.

Terminology	Characteristics
Time-to-contact	Stationary observer and moving object on a collision course
Time-to-passage	Stationary observer and moving object on a crossing course
Time-to-arrival	Moving observer and stationary object/target
Time-to-collision	Moving observer and moving object on a collision course
Time-to-coincidence	Moving object and stationary or moving object on a collision course from the perspective of a distant, uninvolved observer

However, the usage of the terms among researchers can be somewhat confusing and inconsistent. Because of these small differences, which have no impact on the underlying concept, and the lack of a formal taxonomy, the terms are often used interchangeably in literature, even being treated as synonyms (see Bootsma & Oudejans, 1993; Parsonson, Isler, & Hansson, 1996; Schiff & Oldak, 1990; Tresilian, 1991). In addition to the terms listed in Table 1, Hancock and Manser (1998) also suggested the term time-to-go which is supposed to describe a stationary observer facing a moving object on a collision or bypass course. Due to its proximity to time-to-contact (TTC) regarding physical parameters, the use of two different terms appears obsolete. Consequently, the term time-to-go has never been able to establish itself within the framework of urban traffic research, even 20 years after Hancock and Manser introduced their taxonomy, and time-to-go has presumably not been employed again in this research field since Carel’s publication in 1961. Today, the term can be found in aircraft research describing an unrelated metric.

When investigating road-crossing scenarios from the perspective of a pedestrian who is facing approaching vehicles, TTC as a metric meets the required characteristics. Seen from the opposite side, in other words, from the view of the driver approaching the pedestrian, the corresponding metric is termed time-to-arrival (TTA). TTC and TTA can be determined from three-dimensional low-order information, or deduced solely from two-dimensional optical variables, such as looming. The low-order information method builds on the ratio of the oncoming object’s distance to the approaching velocity and is computed as follows:

TTC = \frac{{distance}_{object}}{velocit y_{object}} TTA = \frac{{distance}_{object}}{velocit y_{observer}}

The looming method, on the other hand, consists of Lee’s (1976) concept using the variable

τ

, also noted as tau, relying on the visually perceived angular size of the approaching object, for example, vehicle height, that is taken in proportion to the instantaneous rate of its size expansion—perceived while approaching—and is computed as follows:

τ (t) = \frac{(angular separation of any two image points of the object)}{(rate of separation of the image points)}

While numerous studies have shown that human participants are capable of estimating low-order parameters, such as distance and velocity, humans will mainly rely on the perceived change of optic array to judge an approaching object (Lugtigheid & Welchman, 2011; McLeod & Ross, 1983; Yan, Lorv, Li, & Sun, 2011). Interestingly, this contrasts machine systems such as autonomous vehicles which usually deduce temporal distances based on three-dimensional, sensor-tracked parameters such as described in the low-order information method.

Measurement Methods

Different study concepts have been introduced for the investigation of participants’ judgments of approaching objects, of which an overview is given in Figure 1. Generally, they can be divided into estimation tasks and discrimination tasks.

Figure 1.

Overview of classical tasks evaluating participants’ impact time judgments.

Estimation tasks contain the subsets coincidence anticipation and interceptive action (Tresilian, 1995). The concept of coincidence anticipation requires participants to estimate the collision moment with an approaching object after the object has disappeared from sight by extrapolating the afore perceived motion and estimate the moment when the collision would have taken place, for example, by pressing a button at the respective moment (e.g., Cavallo & Laurent, 1988). The disappearance of the object from the observer’s sight can be due to an occlusion or a virtually induced vanishing. The interceptive action concept requires participants to react to a specific incident by performing a specific movement. The task can, for example, consist in catching an approaching object (e.g., van der Kamp, Savelsbergh, & Smeets, 1997) or avoiding the object in the last possible moment (e.g., Li & Laurent, 1995). Tasks that involve the estimation of absolute numerical values such as the temporal or spatial distance of an approaching vehicle make little sense due to the human innate limitations in making such judgments (see Hills, 1980).

Discrimination tasks use standardized psychophysical techniques aimed at measuring psychometric functions to determine, for example, just-noticeable differences (see Green & Swets, 1966). Those tasks usually require a pairwise comparison leading to determination of a discrimination threshold. Participants may face, for example, two approaching objects, either simultaneously or successively, and judge pairwise which of the two objects will arrive first (e.g., DeLucia, 1991). Regan and Hamstra (1993) used an uncommon discrimination task variant following the psychophysical procedure introduced by McKee (1981), in which the participant was asked to estimate whether a given stimulus was smaller or greater than the mean of the entire set of stimuli, that is, within group.

The within-group comparison remains a rare procedure, while coincidence anticipations, interceptive actions, and pairwise comparisons are the commonly encountered experimental methodologies in this field. When choosing an experimental design, it should be noted that estimation tasks and discrimination tasks look at different aspects: While an estimation task will investigate the perception of the approaching object relative to the observer, the discrimination task will determine the participant’s ability to detect changes or differences. Therefore, when investigating specific variables in an experiment, for example, the performance in a real environment versus a virtual one, the outcome may eventually depend on the chosen task type because the human estimation ability is cognitively dissociated from the human discrimination ability (see Seward, Ashmead, & Bodenheimer, 2007).

Supportive Depth Cues

When an observer is confronted with an approaching object, the perception of the object’s spatial distance and its rate of change are important heuristics. Therefore, it is of little surprise that depth cues and their availability to the observer affect the judgment of impending collisions. Over time, many different depth cues have been determined, investigated, and discussed regarding their availability and effectiveness over spatial distance. Although depth cues have just a supporting function when judging an approaching object, they were given their own section here in which the bases and different types are summarized. This shall sensitize researchers who investigate impending collision judgments—especially in virtual environments—as to which cues may affect the depth perception within a specific range and should therefore be provided to ensure the validity of the findings.

Cues that allow inference of three-dimensional depth information from a two-dimensional still image are termed pictorial depth cues, whereas other cues are categorized as nonpictorial depth cues and are related to motion, the oculomotor system, or stereoscopy. A further distinction can be made by dividing into depth cues that rely either on monocular information or require binocularity. An overview of different depth cues and their effective range, their classification into pictorial and nonpictorial depth cues, and whether they rely on monocular or binocular information is given in Figure 2, with data based on Cutting and Vishton (1995), Nagata (1989), and Renner, Velichkovsky, and Helmert (2013).

Figure 2.

Overview of different depth cues and their approximate effective range.

Pictorial depth cues:

Occlusion: Objects that are closer visually overlap objects farther away, which allows observers to determine the relative but not absolute distance toward them.

Relative size: When objects approach the observer, the object’s relative size increases on the retinal image. Thereby, the observer’s familiarity with the object’s size is of importance, but even when the object is unknown to the observer, smaller objects may appear relatively farther away (Sousa, Brenner, & Smeets, 2011; Sousa, Smeets, & Brenner, 2012).

Relative density: With increasing distance, the relative retinal density of surface texture will increase. However, relative density is merely around the threshold that is considered effective for perceiving depth (Braunstein, 1976; Cutting & Vishton, 1995; Marr, 2010; K. A. Stevens, 1981).

Height in visual field: Objects that have smaller angular distance to the horizon from the observer’s point of view, in other words, objects that are vertically closer to the horizon, appear to be farther away. Theoretically, on a flat surface, an object’s absolute distance can be extracted solely from its height in field, with the effective range lying beyond $2 m$ for an upright observer, although the effectiveness of this cue diminishes curvilinearly over distance (Cutting & Vishton, 1995; Rand, Tarampi, Creem-Regehr, & Thompson, 2011; Sedgwick, 1983).

Aerial perspective: Objects at great distance have lower contrast and lower color saturation, usually involving a bluish cast, due to light scattering by the atmosphere. Aerial perspective is the only depth cue with increasing effectiveness over increasing distance, though the effective range will largely depend on the environmental and meteorological conditions (Cutting & Vishton, 1995; Nagata, 1989). Considering that objects at a great distance become indistinct and that the effective range of aerial perspective usually starts at no lower than a minimum of a few hundred meters, this depth cue is ineffective when judging approaching objects such as vehicles.

Nonpictorial depth cues:

Motion parallax: When an observer is moving, the apparent relative motion of stationary objects against the background gives the observer clues about the relative distance of those objects, with nearby stationary objects having a higher relative motion than those that are farther away. Mathematically, motion parallax can deliver absolute depth information (Ferris, 1972), though effectiveness declines rapidly with increasing distance and eventually becomes ineffective for distances beyond the action space (Cutting & Vishton, 1995). Also, effectiveness is further reduced when the respective objects are in the axis of the observer’s movement, which in application-oriented cases is usually the case, and not lateral to the motion axis.

Accommodation: Kinesthetic sensations when contracting and relaxing ciliary muscles that are responsible for changing focal length deliver cues about the depth of the object on which the observer is focusing. However, accommodation is solely effective for distances less than $2 m$ (Fisher & Ciuffreda, 1988; Okoshi, 1976).

Convergence: Due to binocularity, the two eyeballs are moved inward to create an intersecting focus point. The extraocular muscles that are responsible for this movement deliver kinesthetic sensations, similar to accommodation, and, consequently, cues about the depth of the focus point. Like accommodation, effectiveness is limited to about $2 m$ (Cutting & Vishton, 1995; Nagata, 1989; von Hofsten, 1976), although some sources claim that effectiveness may be applicable as far as $10 m$ (Okoshi, 1976).

Binocular disparity: Due to horizontal separation of the eyes, the eyes perceive two slightly different images of the scene, which are eventually fused into one image. The slight differences between the two original images, however, allow subconscious extraction of depth information by triangulation. The effective range of binocular disparity varies strongly across literature, with investigators reporting different thresholds at which binocular disparity as a depth cue becomes ineffective: Some researchers suggest the cue’s effectiveness to be limited to the personal and action space (<30 m), while others claim an effectiveness of up to $135 m$ (cf. Cavallo & Laurent, 1988; Cutting & Vishton, 1995; Foley, 1991; Lappin, 2014; Nagata, 1989; Palmisano, Gillam, Govan, Allison, & Harris, 2010; Schiff, 1980). With varying interpupillary distances across participants certainly playing a role (discussed in the Human Factors section), it also has to be noted that stereoweakness and stereoblindness are not uncommon in the general population and affect the effective range (Cutting & Vishton, 1995). Clearly, binocular disparity is most effective in near distances with the effect decreasing linearly when distances increase.

Additional depth cues that are also commonly mentioned in literature include texture gradients (e.g., Gibson, 1950), linear perspective (e.g., Kubovy, 1988), light and shading (e.g., Boring, 1942), kinetic depth (e.g., Wallach & O'Connell, 1953), kinetic occlusion (e.g., Kaplan, 1969), and gravity (e.g., Watson, Banks, von Hofsten, & Royden, 1992). These cues have been neglected in the list presented here, as they represent either some combination of several of the cues discussed earlier or are extremely ineffective in their psychological effect of revealing depth (Cavanagh & Leclerc, 1989; Cutting & Millard, 1984; Cutting & Vishton, 1995).

Factors Affecting Impending Collision Judgments

Human Factors

Over recent decades, numerous studies have been conducted on the perception of approaching objects. Participants across those studies all have in common that they have persistently underestimated the temporal distances to approaching objects, that is, TTC or TTA. A meta-analysis visualizes this underestimation effect in Figure 3 and attempts to quantify it, though without taking into account the numerous differing variables across the studies, in other words, ignoring the strongly differing experimental designs, the varying number of participants, and the varying number of assessed estimations (cf. Caird & Hancock, 1994; Cavallo & Laurent, 1988; Hancock & Manser, 1997; Horswill, Helman, Ardiles, & Wann, 2005; Mathieu, Bootsma, Berthelon, & Montagne, 2017; McLeod & Ross, 1983; Petzoldt, 2014; Recarte, Conchillo, & Nunes, 2005; Schiff & Oldak, 1990; Sidaway, Fairweather, Sekiya, & McNitt-Gray, 1996; Tharanathan & DeLucia, 2006). Experiments by Geri, Gray, and Grutzmacher (2010) and by Tharanathan and DeLucia (2006) have indicated that temporal distances were estimated lower when the observer was moving toward the stationary object, thus the TTA, than when the object was moving toward the stationary observer, thus the TTC. Nevertheless, due to the similarity of events, this meta-analysis takes both scenarios into account. An overview is shown in Table 2, indicating whether the experiments investigated

the TTC or

the TTA,

Table 2.

Experimental Characteristics of Studies Examined in the Meta-Analysis.

		McLeod\ and\ Ross\ (1983)	Cavallo\ and\ Laurent\ (1988)	Schiff\ and\ Oldak\ (1990)	Caird\ and\ Hancock\ (1994)	Sidaway\ et al.\ (1996)	Hancock\ and\ Manser\ (1997)	Recarte\ et al.\ (2005)	Horswill\ et al.\ (2005)	Tharanathan\ and\ DeLucia\, (2006)	Petzoldt\ (2014)	Mathieu\ et al.\ (2017)
Mode	Time-to-contact			•	•		•		•	•	•	•
Mode	Time-to-arrival	•	•			•		•		•
Environment	Reality		•					•
	Recorded video	•		•		•		•	•
	Simulation				•		•			•	•	•

Table 3.

Overview of Factors That May Affect the Human Perception of Oncoming Objects.

Human factors	Compositional factors	Technical factors
Erring in safety direction	Looming	Monoscopy/stereoscopy
Gender	Occlusion	Static/dynamic focal plane
Age	Relative size	Field of view
Driving experience	Relative density	Brightness
Risk tolerance	Height in visual field	Contrast
Physiological characteristics	Motion parallax	Texture
	Approaching angle	Abstractness of virtual events
	Approaching velocity
	Spatial distance
	Temporal distance
	Observation time
	Light and weather conditions

and also whether the experiments

conducted the study in a real environment,

displayed a previously recorded video of a real traffic scenario, or

used a simulated scenario.

Please note that in order to facilitate the reading flow, the research overview in this section and also the sections hereinafter use the term TTC, referring eventually also to studies that investigated the TTA.

Figure 3.

Meta-analysis of impending collision judgments from an egocentric perspective, across 11 different studies that have used contextual stimuli, in other words, realistic environments containing depth cues, plotted as estimations relative to the actual temporal distances, that is, TTCs and TTAs.

Despite the fact that all of the analyzed studies contained a lifelike environment, the designs of those experiments varied strongly and are thus difficult to compare. Nevertheless, this meta-analysis allows an approximate idea on how to globally grade the ratio of TTC estimation to the actual TTC. A linear regression (R² = $.89$ ) indicates a ratio of about 75% with the slope having an SE = $0.03$ . Schiff and Oldak (1990) attributed this phenomenon to the biological instinct of self-preservation with built-in or learned tendency to err in the direction of safety in potentially hazardous situations, particularly when the perceptual response system is lacking precision. This is supported by the observation that those underestimations occur in studies that have used contextual stimuli, for example, representations of approaching vehicles and sophisticated backgrounds. Other studies that have used highly simplified stimuli, for example, an approaching square on a plain white background, varied strongly, indicating partially substantial underestimations (Carel, 1961; Schiff & Detwiler, 1979; Todd, 1981), slight overestimations (Gray & Regan, 1998; Kaiser & Mowafy, 1993), as well as large overestimations of the TTC (DeLucia, Preddy, & Oberfeld, 2016; Geri et al., 2010).

When looking on gender as a between-subjects factor, significant differences were found, indicating lower TTC values for female than for male participants (see Caird & Hancock, 1994; Manser & Hancock, 1996; McLeod & Ross, 1983; Schiff & Oldak, 1990). A possible explanation may lie in women’s pursuit for greater safety margins (see Hills, 1980; Kadali & Vedagiri, 2012; Konečni, Ebbeson, & Konečni, 1976; Montgomery, Kusano, & Gabler, 2014; Parsonson et al., 1996) and thus an enhanced underestimation of the TTC, whereas the higher risk tolerance of men by implication leads to higher accuracy of the male TTC estimation. In addition, numerous experiments in the past have revealed that men tend to have fewer difficulties with regard to solving spatial tasks when compared with women: A meta-analysis by Voyer, Voyer, and Bryden (1995) reviewing 190 experiments with regard to spatial visualization, spatial perception, and mental rotation showed a significant gender difference with 112 of the experiments in favor of men and 3 experiments in favor of women, while no significant difference was observed in 75 of those experiments. Furthermore, men account for a larger share of traffic, resulting in increased driving experience (see Kirkham & Landauer, 1985; McGuckin & Fucci, 2018; Polus, Hocherman, & Efrat, 1988). In this light, it should be noted that an increased driving experience leads to a more accurate TTC estimate as a male-only study by Cavallo and Laurent (1988) has shown. Interestingly, Recarte et al. (2005) found no significant difference between the TTC estimations of men and women in their study with a sample of fairly young participants ( $\emptyset 23.5 years$ , SD = $2.5$ ) and similar driving experiences between both genders ( $\emptyset 24, 355 km$ for men, $\emptyset 27, 101 km$ for women). This could support the thesis that driving experience might to a certain extent have a larger impact on the estimation than the gender itself, at least when it comes to young subjects.

The participants’ age also appears to influence the accuracy of TTC estimations: Schiff, Oldak, and Shah (1992) observed that elder participants tend to underestimate the TTC to a greater extent than their younger comparison group. This effect was particularly noticeable when comparing female participants only. The differences between younger male and elder male subjects were less significant, which has a certain consistency with Matthews’ (1986) findings of attributing elder male drivers with an apparent overconfidence in traffic. A very similar age effect, as well as age × gender interaction, as reported by Schiff et al. (1992), was observed by Hancock and Manser (1997), and also Andersen and Enriquez (2006), as well as Dommes, Cavallo, and Oxley (2013), observed a strong deterioration of adults' TTC judgments with increasing age. Transferred to a practical application, age has been shown to affect road-crossing choices and may be partially related to difficulties of elderly people and children in judging approaching vehicles (see Barton & Schwebel, 2007; Connelly, Conaglen, Parsonson, & Isler, 1998; Demetre et al., 1992, 1993; Dommes & Cavallo, 2011, 2012; Dommes, Cavallo, Dubuisson, Tournier, & Vienne, 2014; Dommes et al., 2013, 2015; Dommes, Cavallo, Vienne, & Aillerie, 2012; Hills, 1980; Lee, Young, & McLaughlin, 1984; Lobjois & Cavallo, 2007, 2009; O'Neal et al., 2018; Oxley, Ihsen, Fildes, Charlton, & Day, 2005; Staplin, 1995; Young & Lee, 1987).

Physiological characteristics can also affect motion and depth perception: Banister and Blackburn (1931) measured interpupillary distances of 258 students and classified them according to their performance at ball games. The analysis of that data revealed a correlation between larger eye distance and higher performance at ball games. This was explained with the better stereoscopic vision that results from the wider distance between the eyes, increasing the effectiveness of depth cues inferred from binocular disparity and convergence (see the Supportive Depth Cues section). Indeed, binocular disparity as a contributing cue for judging motion-in-depth and speed-in-depth has been demonstrated to be effective in various experiments (González, Allison, Ono, & Vinnikov, 2010; Khuu, Lee, & Hayes, 2010).

Compositional Factors

Although some researchers (e.g., Caird & Hancock, 1994; Hancock & Manser, 1997; Sidaway et al., 1996) described that certain factors lead to a “greater accuracy” of the TTC estimation, it should be noted that due to the consistent underestimation discussed in the Human Factors section, compositional factors that lead to a higher estimation of the temporal distance also increase the accuracy by implication. Therefore, it is more appropriate to classify possible effects as increase/decrease of the relative estimation because some increasing factors may—with sufficient effect—lead to an exceedance of the actual TTC and by that lower accuracy itself again.

Todd (1981), as well as Kaiser and Mowafy (1993), showed that even when lacking spatial information, it is possible to estimate the TTC by solely relying on the perceived looming of objects. However, Todd (1981) also stated that sensitivity to observe accelerations under those conditions turned out to be extremely poor. There are many other factors and heuristics, for example, depth cues, that can influence a person’s perceptual process. Heuristics such as depth cues can support the perception and estimation process especially in case of restricted availability of invariants, such as looming, that may result from sensory or cognitive limitations, such as in case of low contrast (DeLucia, 2004). Various studies have sought to investigate the influence of depth cues on perception of approaching objects. Pictorial depth cues from occlusion, relative size, relative density, and height in visual field have proved to have a significant impact on TTC estimations (DeLucia, 2004; DeLucia, Kaiser, Bush, Meyer, & Sweet, 2003; Vincent & Regan, 1997). Depth information from the aerial perspective, which is effective at a great distance only, may be neglected though when setting up an experimental environment (see the Supportive Depth Cues section).

Special attention has been given to the relative size of an approaching object, with multiple studies suggesting that this cue can overrule perceived looming ( $τ$ ), termed the size-arrival effect. DeLucia (1991), DeLucia and Warren (1994), Kappé and Korteling (1995), Stewart, Cudworth, and Lishman (1993), as well as DeLucia (1999) have demonstrated this effect by investigating the estimated impact moment with different-sized same-type objects, for example, squares displayed on a screen, with virtual self-motion toward the object (i.e., TTA) or virtual approach of the object toward the participant (i.e., TTC). Under identical approaching parameters, participants estimated the respective impact time for the larger objects to be more imminent than for the smaller ones. Van der Kamp et al. (1997) conducted a comparable experiment in a real environment: Four luminous balls of different diameters were approaching the participants in the dark. Interestingly, van der Kamp et al. observed the size-arrival effect solely in monocular viewing conditions and not binocularly. Caird and Hancock (1994), Horswill et al. (2005), as well as Mathieu et al. (2017) looked more application-oriented into the size-arrival effect, comparing the TTC estimations for small motorcycles, large motorcycles, compact cars, full-size cars, and vans. The participant was facing the approaching vehicles displayed on a screen that were either being rendered with simulation software (Caird & Hancock, 1994; Mathieu et al., 2017) or recorded in real traffic environments beforehand (Horswill et al., 2005). In all three studies, a significant effect was observed with larger vehicles being estimated to be temporally closer compared with smaller ones.

When looking on relative density of surface texture, Vincent and Regan (1997) noticed that a mismatching expansion of the approaching object’s surface texture opposed to looming, in other words the object’s outline expansion, affected TTC estimations significantly, even if the mismatching expansion error was held as low as 10%.

In addition to pictorial depth cues, the nonpictorial depth cue motion parallax was also found to be effective (DeLucia et al., 2003). Further nonpictorial depth cues are discussed in the Technical Factors section, as cues related to a shifting focal plane and binocularity are more of a technical challenge and not related to environmental composition.

Manser and Hancock (1996) investigated the influence of the trajectory angle of an approaching object and concluded that a larger angle ( $40°$ ) results in a lower TTC estimation when compared with a straight oncoming trajectory ( $0°$ ). Hence, they argued that peripheral vision is less efficient in terms of extracting information regarding radial optical flow. This is supported by the findings of Regan and Vincent (1995) who observed increased difficulties to judge approaching objects in peripheral view when compared with those approaching in the foveal visual field.

Caird and Hancock (1994) investigated spatial distance as a variable and observed that a larger distance ( $61 m$ ) of an approaching vehicle results in an increased relative estimation of the TTC in comparison with a closer distance ( $30.5 m$ ). This is consistent with Petzoldt’s (2014) findings, according to which a higher velocity ( $50 km / hr$ ) of an approaching vehicle compared with a lower one ( $30 km / hr$ ) leads to a higher TTC estimation as well: For a constant TTC, a higher velocity signifies a larger distance, as those parameters are inseparably connected in a triangular physical relation ( $velocity \times TTC = distance$ ). The same effect was also observed when the virtual environment was strongly simplified, as Kappé’s and Korteling’s (1995) experiment demonstrated. Again, an explanation can be that the larger distance may appear to be psychologically less of an imminent threat and thus trigger less need for a subconscious safety margin, discussed in the Human Factors section. Opposite to those findings stand the results of Cavallo’s and Laurent’s (1988) study conducted in a real environment, which revealed no significant differences between estimations made for lower ( $30 km / hr$ ) and higher ( $90 km / hr$ ) velocities, unless visually impoverished conditions were included in form of restricted visual field or monocular vision.

The chosen temporal distance itself also plays a role for the capacity to estimate the TTC of oncoming objects. Schiff and Detwiler (1979), who analyzed a variety of TTC estimations covering 2 s, 4 s, 6 s, 8 s, 10 s, and 16 s, concluded that there is little point in assessing TTC estimations above 10 s since participants do not seem capable of exploiting perceived information beyond that point. McLeod and Ross (1983), as well as Thomson (1983), deduced an even lower limit of about 8 s, up to which consistent estimations are to be expected. These suggested thresholds are not firm but rather approximate values beyond which estimation quality was observed to deteriorate rapidly. Furthermore, Schiff and Detwiler (1979) state that the observer’s critical perceptual-motor adjustments when facing an approaching object occur within the last 4 s prior to contact anyhow.

Studies that examined the effect of observation time on TTC estimations have achieved divided conclusions. While McLeod and Ross (1983), who compared observation times of 2s, 3 s, 4 s, 5 s, and 6 s, found no significant differences regarding estimation accuracy of the TTC, Manser and Hancock (1996) argued in their study that the duration of vehicle observation has an influence on TTC estimation. However, it should be noted that Manser’s and Hancock’s observation times— $8.92 s$ , $10.42 s$ , and $11.92 s$ —were not dissociated from different spatial distances— $80 m$ , $60 m$ , and $40 m$ , respectively—making it more likely to attribute the effect to the different spatial distances than to the small differences in observation time. Yet, Groeger and Cavallo (1991), who analyzed observation times of 2 s and 6 s as an individual variable, found significant higher and thus more accurate TTC estimations for the longer observation time than for the short one.

Besides all those factors that rather depend on depth cues and parameters of the study design, environmental conditions such as the weather can also have an impact: Snowden, Stimpson, and Ruddle (1998), who compared self-perceived speed of drivers in clear, misty, and foggy conditions, noticed that subjects underestimated their speed significantly when visibility dropped. De Bellis, Schulte-Mecklenbeck, Brucks, Herrmann, and Hertwig (2018), as well as Gegenfurtner, Mayser, and Sharpe (1999), noted that a decreased environmental brightness may lead to an underestimation of driven speed.

Technical Factors

Because most of the studies investigating TTC estimations are carried out with some virtual device or screen, there are technical factors that may influence the perception and should therefore be considered.

The importance of a stereoscopic view for three-dimensional tasks has been already known for centuries, as the observation by Molyneux in 1690 shows:

And as a conclusion to the whole shall only add one experiment that demonstrates we see with both eyes at once; and 'tis, that which is commonly known and practised in all tennis-courts, that the best player in the world hoodwinking one eye shall be beaten by the greatest bungler that ever handled a racket; unless he be used to the trick, and then by custom he gets an habit of using one eye only. (pp. 293–294)

With humans being used to stereoscopic images in everyday life, the reduction to monoscopic depth information will—unsurprisingly—affect their performance in judging their three-dimensional surrounding space. Various studies sought to compare the perception of approaching objects when provided with monocular information as opposed to binocular one (e.g., Cavallo & Laurent, 1988; Gray & Regan, 1998; van der Kamp et al., 1997). For these experiments, participants observe the three-dimensional scenery partly monocularly and partly binocularly, with one eye being eventually suppressed for the monocular condition. Alternatively, the effect can also be achieved by just switching between a monoscopic and stereoscopic visual representation of the scenery while participants maintain a binocular view continuously: A monocular view and a monoscopic display both result in a lack of binocular depth cues, such as binocular disparity and convergence. While a stereoscopic head-mounted display (HMD) will provide binocular cues, depth cues resulting from accommodation will still be lacking because the focal plane remains invariant in current state-of-the-art HMDs, though the effectiveness of accommodation is limited to very close distance anyhow (see the Supportive Depth Cues section). This should be considered for experiments with close acting ranges, such as surgical simulations (see Satava & Jones, 2003). Overall, participants performed better when being provided with binocular information, though Cavallo and Laurent (1988) observed an advantage predominantly at closer distances, suggesting that the effectiveness of binocular disparity was confined to about 75 m in their study. In the case of a small approaching object within the range of a few meters, Gray and Regan (1998) found TTC estimations primarily to rely on binocular cues, concluding that estimation accuracy may significantly improve when both monocular and binocular information is available to the observer. Van der Kamp et al. (1997) also observed the superiority of TTC estimations when binocularity was added. They demonstrated that in the case of binocular vision, lacking information about the relative size of approaching objects (see the Compositional Factors section) became irrelevant—at least within the short range of 2 m that was investigated in their study.

For now, HMDs provide solely a reduced field of view (FOV), which can affect visual perception. Cavallo and Laurent (1988) compared participants’ impact time estimations when heading toward a stationary object while having a normal visual field and while having a reduced FOV limited to the foveal and parafoveal visual field of about $10°$ . Estimations with full visual field turned out to be significantly more accurate. However, as a post hoc analysis revealed, the difference was highly significant for inexperienced drivers, whereas no significant difference was observed for experienced ones, although a tendency was also apparent in the latter case (p = $.051$ ).

From a technological perspective, the brightness and contrast of the display may have an effect on the perception of velocities: Takeuchi and De Valois (2000) investigated the brightness as a factor when judging velocities of objects displayed on a screen and observed that a lower brightness may lead to a reduced capability of perceiving a differential between two different velocities. Anstis (2003), Blakemore and Snowden (1999), Stone and Thompson (1992), and Thompson (1982) manipulated the contrast of the visual display in their studies and showed that a reduced contrast typically evokes the perception of the object moving at a lower speed.

A large number of studies investigated TTC estimations by making the approaching object virtually vanish at a predefined moment, asking the participants to estimate at what moment the object would have made contact with them by extrapolating the aforeseen motion (see the Measurement Methods section). Because the event of a suddenly disappearing vehicle is not very lifelike, Hancock and Manser (1997) compared virtual scenarios in which the approaching vehicle either vanished “magically” into thin air or disappeared through occlusion by driving behind a bush. The results indicated that participants had significantly greater estimation accuracy in the case of occlusion compared with the vanishing mode.

Some researchers wondered about the influence of texture integrated into the simulation environment. López-Moliner, Brenner, and Smeets (2007) examined the effect of surface characteristics of an approaching object on the TTC estimation within a simplistic simulation and concluded that no significant differences could be observed, no matter whether the surface was textured or not. DeLucia et al. (2003) also argued that globally a richer visual texture of the object or the background surface do not influence TTC estimation performance. Li and Laurent (1995) confirmed with their experiment conducted in a real environment that an increased texture of an approaching ball did not influence the temporal distance, that is, TTC, at which the participant decided to dodge the ball. However, they observed a significant speed increase of the participant’s evasive movement when the ball held a texture compared with when the ball surface was left blank. Studies that observed influences of texture on TTC estimations involved intentional deficiencies in visual representation, such as a mismatching expansion of the surface texture opposed to the perceived looming of an approaching object (Vincent & Regan, 1997), or an insufficient contrast that could be compensated by the presence of a textured background (Blakemore & Snowden, 2000).

Discussion

Influencing Factors

In conclusion, a combination of many factors is incorporated into the judgment of TTCs. An overview of factors discussed throughout the Factors Affecting Impending Collision Judgments section is given in Table 3.

The research field related to the perception of oncoming objects was stimulated by Lee’s concept of optical flow and looming introduced in 1976 and the controversial hypothesis that human TTC estimations rely solely on this monocular cue. Studies that provide participants with solely two-dimensional information may conclude that TTC estimations are made uniquely based on them. However, it is a misconception to believe that other factors and stimuli—once available—do not weigh in. During the 1980s and 1990s, a wave of experiments followed, aiming at identifying factors that may or may not influence TTC judgments. Numerous factors connected to human individuality, study design, experimental environment, and technological aspects were successfully identified and can be partially linked to each other. However, with recent evolution in the technological field and also in regard to human factors, for example, women participating more in traffic than three decades ago, a new investigation of some of these factors may be of interest.

Substantial differences were observed regarding TTC estimations when the representation of the scenario was realistic and contextual compared with abstract experiments that involved simplistic environments. While the results of the realistic experiments varied fairly little with a consistent underestimation of about 75% of the actual TTC (see the Human Factors section), the simplistic studies covered average estimations ranging from 50% (e.g., Schiff & Detwiler, 1979) up to 200% (e.g., DeLucia et al., 2003) of the actual TTC. Although quality variation of virtual textures showed no influence on TTC estimations (see the Technical Factors section), the observation above indicates importance of an overall lifelike environment to achieve realistic results.

When investigating human perception and behavior, human factors will always play a role due to human individuality. Those factors may be invariant (such as gender), mutational (such as age or some physiological characteristics), or trainable (such as driving experience or risk tolerance). However, those factors are not always dissociated from each other: For example, driving experience and risk tolerance are entangled with both gender and age. However, since in the study reported by Recarte et al. (2005) no gender-specific differences could be observed when men and women had the same driving experience, one has to ask if gender can be considered as a factor at all and not merely as an indicator of driving experience and risk tolerance. In the light of the steady decrease in the gap between male and female driving experiences over recent decades (see McGuckin & Fucci, 2018), findings of numerous studies from the 1980s and 1990s may be outdated and a revaluation of gender-specific differences is encouraged.

Compositional factors may support (or also corrupt) the participants’ perception of their environment and as a result affect the given tasks. These factors may be related to the perception of dynamic changes (such as looming), the perception of depth (such as occlusion, relative size, relative density, height in visual field, and motion parallax), or simply to the experimental design (such as approaching angle, approaching velocity, spatial distance, temporal distance, and observation time). When the experiment is conducted outside in a real environment, changing light and weather conditions may also affect the experimental outcome. Compositional factors may interact and support one another and especially compensate for various deficient invariants and heuristics.

Human beings experience limitations when it comes to the extent of the temporal distances up to which reliable estimations can be expected. While Schiff and Detwiler (1979) noted a limitation of $10 s$ and McLeod and Ross (1983) as well as Thomson (1983) set the limit at about $8 s$ (see the Compositional Factors section), a strong inconsistency of estimations can already be observed at around $7 s$ in Caird’s and Hancock’s (1994) experiment. In general, the longer the TTC is set, the wider the variance of estimations is to be expected and the estimations fall into a certain arbitrariness beyond a certain point.

There are technical factors that impede an experimental scenario within a virtual environment to be perceived identically as it would have been within a real environment. Examples are monoscopic displays instead of stereoscopic ones that involve binocular depth information, a static focal plane instead of a dynamic one, as well as a reduced FOV. Furthermore, some attention should be given to the technical settings of the visual display. Parallels can thereby be drawn between the observed effects of poor visual conditions in a real environment, for example, darkness or fog (see the Compositional Factors section), and those observed due to certain technical parameters of the visual display, for example, reduced brightness or contrast (see the Technical Factors section), that evoke the same perceptual mechanisms regarding TTC estimations.

Many of the different factors are strongly intertwined with each other. For example, the effect of a given FOV on TTC estimations may depend on participants’ driving experiences, same as the impact of an approaching object’s relative size or its velocity on TTC estimations will eventually depend on whether the display is monoscopic or stereoscopic.

In summary, the complexity and quantity of influencing factors warrant further research to determine in what way different factors interact with each other and what importance has to be attributed to each of them.

Comparison of Real and Virtual Environments

Between-studies comparison

Virtual environments as a tool represent a mixed bag for perceptual research: Even though they are a valuable tool to investigate scenarios that could not have been easily investigated otherwise, they also possess the potential to distort the collected data. Most of the studies discussed across the Factors Affecting Impending Collision Judgments section were conducted using some display system or VR device, without these setups having been validated against real environments. Researchers may observe the same effect over and over again when investigating a specific phenomenon, without taking into account that the effect may be related to the design of the experimental setup. The size-arrival effect, for example, with larger objects leading to lower TTC estimations (see the Compositional Factors section), was observed by Caird and Hancock (1994), DeLucia (1991, 1999), DeLucia and Warren (1994), Horswill et al. (2005), Kappé and Korteling (1995), Stewart et al. (1993), as well as Mathieu et al. (2017). Given these numerous experiments with varying study designs and parameters, one might believe this effect to be sufficiently documented. However, it should be noted that what all these experiments have in common is they all used some sort of display with one single focal plane in close proximity to the participant, providing solely monocular information. Van der Kamp et al. (1997) conducted an experiment in a real environment with a study design close to some of the ones mentioned earlier. However, because this experiment was conducted in a real environment and no screen was involved, binocular information was available to the participants and, furthermore, the focal plane was shifting with the approaching object. While the size-arrival effect was likewise observed when the participant’s view was artificially reduced to monocular vision, a binocular viewing condition abolished this effect, emphasizing the importance of stereoscopic vision delivering binocular depth cues for this kind of task. Results of experiments by DeLucia (2005), in which participants were exposed to monoscopic as well as stereoscopic displays, also strongly suggest that binocularity may abolish the size-arrival effect.

Contrary to Caird and Hancock (1994), Kappé and Korteling (1995), as well as Petzoldt (2014) who conducted their studies in a virtual environment using a monoscopic display, Cavallo and Laurent (1988) found no influence of velocity on impact time estimations in a real environment (see the Compositional Factors section). Even though Cavallo and Laurent were investigating TTA while the other three experiments studied TTC primarily—Kappé’s and Korteling’s study involved both TTC and TTA measures, although an interaction effect was not investigated—the difference can still presumably be traced to the technological characteristics of the setups: Under impoverished visual conditions, that is, reduced FOV or lacking binocularity, Cavallo and Laurent reproduced similar effects as those observed in the three studies that took place in a virtual environment.

Setups using displays for visualization of the experimental scenario often provide a reduced FOV that is in general significantly smaller than the human far peripheral vision can handle. The experiment conducted by Cavallo and Laurent (1988) in a real environment comparing normal viewing conditions to an artificially reduced FOV ( $10°$ ) revealed differences in TTA estimations (see the Technical Factors section). A post hoc analysis indicated the differences to be solely significant for a group of inexperienced drivers, and not for an experienced one. This shows that differing sources of information may be employed when preconditions differ. Subjects can also rely on nonapparent information sources to fulfill their task as Cavallo’s and Laurent’s experiment demonstrated: Participants sitting in a vehicle and approaching head-on a stationary object within their central visual field also processed visual information from the periphery. Although it is conclusive in this case that participants may involve the peripheral visual flow to deduce speed (see Warren & Hannon, 1988), there are likely research scenarios in which it may be very complex to identify all influencing factors beforehand. If some of these factors are not correctly incorporated into the experimental setup, it will complicate transfer of the study’s findings to real environments. The current example of Cavallo’s and Laurent’s experiment illustrates how laborious the validation process of an experimental setup is: Although a setup could have been possibly validated with a group of experienced drivers and no significance would have been found regarding the estimations with different visual fields, a small variation of a parameter such as the change of the sample’s driving experience could require a new validation of the setup.

Within-studies comparison

Recarte et al. (2005) conducted an experiment in which participants were asked to estimate the impending collision from inside of a vehicle in a real environment as well as while sitting in front of a screen watching recorded footage of scenarios from the same point of view and perspective. They noted that estimations made in reality had a lower variance and also a higher correlation with the actual time than estimations based on the recorded videos, which was observed for all 16 different experimental conditions emerging from combinations of four different speeds ( $60 km / hr$ , $80 km / hr$ , $100 km / hr$ , $120 km / hr$ ) and four different distances ( $75 m$ , $100 m$ , $125 m$ , $150 m$ ). While a higher similarity of experimental distances led to a higher correlation of the estimates within the respective environment in both environmental conditions, a higher similarity of experimental speeds led to a higher correlation of the estimates only in the real environment and not in the virtual one. This suggests that participants experienced more difficulties in differentiating between velocities than in processing different distances when visualized on a screen, as the cross-comparison with the reality setting indicates. Despite the efforts to provide the participants with comparable experimental settings and parameters in both environments, some differences regarding the estimations were evident, suggesting that information sources in both environments were employed differently. Also, an apparent access to additional information sources in reality that were simply not available in the screen setup seems plausible. Despite the highlighted differences between the estimations in both environments, it should nevertheless be mentioned that overall, there was also a great comparability of the results in many other characteristics.

One of the very few reported examples in which a VR setup for research and training purposes was actually subject to a validation study against reality was the experiment conducted by Schwebel, Gaines, and Severson (2008): They designed a moderately complex pedestrian simulator in which the participant stands on an artificial physical curb and faces the scenery of a two-lane bidirectional virtual street. The scenery, which contained cars approaching from both sides, was displayed on three monitors that were arranged in semicircular alignment toward the user. The participant was asked to initiate the street crossing by stepping off the curb whenever feeling safe to do so and with that, stepping onto a pressure plate. This triggered the scenery to morph from first-person view to third-person view, and the participant observed an avatar cross the street at a constant personalized speed that was measured beforehand and adjusted for each participant individually. Schwebel et al. conducted an experiment with 74 adults and 102 children that aimed at comparing the behavior in the simulator to real street crossings and ultimately validate the setup as a tool dedicated to train children toward a safe street crossing behavior. In the experiment, children were asked to perform three different types of tasks: initiating virtual road crossings within the described simulator setup, verbally indicating road crossings facing a real road (shout task), and indicating road crossings equally facing a real road but by doing the first two steps only (two-step task). Adults had a fourth one in addition to these three tasks, in which they actually crossed the real road at their own discretion.

Schwebel et al. (2008) argued that a construct validity was demonstrated through a significant correlation between the behavior in the real and the virtual environment as well as through developmental differences between adults and children observed in both environments. Correlations of parent-reported child temperament and the child’s crossing behavior revealed a convergent validity, while a face validity of the simulator was attested due to the participants’ self-reported perceptions of realism.

However, to conclude that the simulator is fully validated may be hasty, despite the numerous parallels shown between both environments. Although developmental differences, behavioral patterns, and a realistic impression of the virtual environment are all highly important factors in an overall validation of the setup, they do not provide quantitative information about the perceptual similarity between the real and the virtual environment. For this, Schwebel et al. analyzed two variables: for adults, the safety gap, that is, the TTC after the participant successfully crossed the street, and for children, the start delay, that is, the time elapsed between the last vehicle passing and the initiation of the crossing itself.

The generated safety gap may indeed give insight about the pedestrian’s perception of approaching vehicles. However, it must be noted that the experimental differences in the study design between the real and the virtual environment add additional variables and uncertainties: While in reality, participants were capable of adjusting their walking speed by continuously revaluating the approaching vehicle while crossing and thus speed up or slow down, the crossing speed in the simulator was rigid and visualized by an avatar without the user being capable of influencing the speed during the crossing process.

The start delay analyzed for the children, on the other hand, provides very little insight about their effective perception of approaching vehicles. This metric serves predominantly as a proxy measure of the cognitive processing time for judging the next approaching vehicle before initiating the crossing (Thomson et al., 2005). The importance of this metric is beyond question: Plumert, Kearney, and Cremer (2004) demonstrated that children show a riskier crossing behavior by having a greater start delay than adults, despite accepting the same gap in between two passing vehicles. Nevertheless, this metric cannot reveal how the pedestrian perceives and estimates the approaching vehicle’s distance and velocity.

Most important, it should be pointed out that the comparison of these two variables—safety gap and start delay—conducted by Schwebel et al. (2008), investigated the correlation between the real and the virtual environment. This means that a significant correlation between the two environments provides minimal information, giving insight solely into the similarity of the tendency in both environments. For example, in the case of a significant correlation, a participant with a shorter safety gap in the real environment will most likely also have a shorter safety gap in the virtual one; the significant correlation will not tell whether the values are identical nor even reveal the relativity to each other. Simulators that underwent such a validation process would allow solely nominal or ordinal measurement scales for research questions (see S. S. Stevens, 1946). For all of these reasons and despite the impressive extent of the study by Schwebel et al., it would be precipitate to declare an overall validity of the setup solely based on these observations.

David C. Schwebel (University of Alabama at Birmingham, USA) kindly provided the raw data collected in his study dating over a decade ago to the author of the work presented here to allow a more in-depth comparison between the road crossings using the VR setup and those observed in reality. The goal of the further analysis was to determine whether there are significant differences—despite the observed correlation—in the participants’ behaviors when crossing the road in reality and when simulating the crossing in that specific VR setup. Because only adult participants effectively crossed the road during the experiment and the sample of children skipped this task, the analysis was reduced to the adult sample. Two participants were excluded due to incomplete data, making the reanalyzed sample consist of 72 participants in total. In a first step, the scenario parameters in the real environment were compared with those in the virtual one: the density of the traffic and the participants’ road-crossing times. Although the virtual road-crossing scenario displayed a replicable number of cars per minute, the scenario in reality was somewhat difficult to control and was owed to the situational chance. Eventually, the virtual crossing scenario contained on average $11.95 cars / \min$ (SD = $2.06$ ), while in the real world participants faced on average $13.9 cars / \min$ (SD = $2.32$ ). A paired-samples t-test confirmed a highly significant difference regarding the car density between the real and the virtual environment ( $t (71) = 5.271$ , p $< .001$ , r = $.53$ ). The participants’ road-crossing times also revealed highly significant differences between the two environments: The crossing duration in the virtual environment that was individually determined for the participants before the experiment was on average 5.5 s (SD = 0.93), while the crossing duration in the real environment remained variable during the crossing itself and was on average $4.73 s$ (SD $= 0.67$ ). Because the requirements for normal distribution (see Field, 2013) were not fulfilled for the crossing times, the samples were analyzed via a Wilcoxon signed-rank test (T $= 108.5$ , p $< .001$ , r = $.56$ ). The differences in road-crossing time and traffic density between the real and the virtual environment—visualized with box plots in Figure 4—as well as the fact that road-crossing speeds remained constant in the virtual environment while variable and adaptable to the current road situation in the real environment, complicate a proper comparison between the two environments.

Figure 4.

Box-and-whisker plots visualize the distribution of participants’ road-crossing durations (left) and the traffic density the participants were confronted with during the experiment (right). Box-and-whisker plots structure: Values between the lower and upper quartile are represented by the box, while the whiskers identify estimations within 1.5 IQR of the lower and upper quartile. Outliers are plotted with lozenges. The horizontal line in the box portrays the median, and the small square shows the arithmetic mean value.

The participants’ average remaining time to the approaching car after crossing the road—that is, the safety gap measured as TTC—amounts to $8.95 s$ (SD $= 2.7$ ) for the real road-crossing, whereas in the simulated crossings, the safety gap was at $4.97 s$ (SD $= 1.26$ ) only. A paired-samples t-test confirmed the highly significant difference regarding the safety gap between the two environments (t $(71) = 11.967$ , p $< .001$ , r $= .82$ ). Considering that the road crossing time itself differed significantly between the two environments, a further look was given to TTCs at the crossing initiation, neglecting the subsequent crossing durations. Yet, also in this case, the difference between the real environment ( $\emptyset 13.67 s$ , SD $= 2.66$ ) and the virtual one ( $\emptyset 10.47 s$ , SD $= 1.78$ ) remained highly significant (t $(71) = 8.956$ , p $< .001$ , r $= .73$ ). The distribution of the safety gaps at crossing initiation and those at crossing completion are visualized via box plots in Figure 5.

Figure 5.

Box-and-whisker plots visualize the distribution of the TTC at crossing initiation (left) and after crossing completion (right). For the structure of the box-and-whisker plots, see Figure 5.

Further attention was given to the start delay that turned out to be on average a bit higher in the virtual setting ( $\emptyset 0.84 s$ , $SD = 0.36$ ) compared with the real road crossings ( $\emptyset 0.58 s$ , SD $= 0.43$ ). A Wilcoxon signed-rank test suggests a highly significant difference between the two samples (T $= 1888.5$ , p $< .001$ , r $= .29$ ). Of particular note are some occurring negative start delays in the real road-crossing scenario, whereas all of the start delays in the virtual setting were positive (see Figure 6). The negative start delay values signify that participants in the real scenario at times initiated the crossing, that is, stepping onto the street, before the last car had entirely passed, while in the virtual environment, the participants always waited until the virtual car had entirely crossed.

Figure 6.

Box-and-whisker plots visualize the distribution of participants’ start delay (left) and the chosen gap size (right). For the structure of the box-and-whisker plots, see Figure 6.

Finally, the gap size chosen by the participants to cross in between two vehicles was compared: At the real road crossings, participants chose on average gap sizes between cars of $14.65 s$ (SD $= 3.13$ ), while in the virtual road crossings, the average gap size amounted to only $9.72 s$ (SD = $1.26$ ). A paired-samples t-test confirmed a highly significant difference (t $(71) = 14.767$ , p $< .001$ , r $= .87$ ) which is also visualized with box plots in Figure 6.

In conclusion, it must be noted that the VR setup, introduced and examined by Schwebel et al. in 2008, revealed, after all, some essential and highly significant differences compared with real-world behavior. Participants demonstrated a much riskier behavior in the virtual environment by choosing significantly smaller gap sizes and leaving significantly smaller safety gaps when crossing the road. This suggests that participants perceive and process approaching vehicles in virtual environments differently to those in real environments. Eventually, this can be due to a differing perception of the vehicles’ approachment in virtual environments or also due to experiencing a different safety feeling in virtual environments, not perceiving the virtual imminent threat in the same way as in real environments. In addition, the lower—partly negative—start delay at real road crossings suggests that participants act differently—some might say more efficiently—in their familiar environment.

This VR setup example demonstrates that even allegedly validated simulators may bear essential differences to real-world settings and points out that published results of experiments carried out with various VR setups that have not been thoroughly validated beforehand have to be handled with care. This does not require the discarding of published results but shall increase the researchers’ sensitivity when interpreting those findings. This shall also not question the use of VR setups for perceptual and behavioral experiments, as these setups come with tremendous advantages and open up whole new research possibilities (see Scarfe & Glennerster, 2015), but it shall emphasize the importance of the validation necessity for VR setups that are intended to be used in behavioral research. Thereby, the validation process shall be customized to the intended research or education purposes. The degree of similarity regarding perception and behavior between real and virtual environments will indicate which measurement scale may be applicable for research questions (see S. S. Stevens, 1946). So even with significant differences between both environments, a virtual environment with suitable arrangements may still be a valuable tool for behavioral research or be validly applied for educational purposes, for example, for evolving children’s cognitive skills in traffic environments (e.g., Demetre et al., 1993; Schwebel et al., 2008).

Footnotes

Acknowledgements

The author would like to express his gratitude to his colleague Georg N. Dyszak for his constant participation and meaningful discussions related to the work presented here. Furthermore, with this review building upon numerous extrinsic experiments, the author would like to thank the many researchers who took the time to enter fruitful discussions about their original work.

Declaration of Conflicting Interests

The author(s) declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.

Funding

The author(s) disclosed receipt of the following financial support for the research, authorship, and/or publication of this article: This research was funded by the Fulbright Program and the Studienstiftung des Deutschen Volkes.

ORCID iD

Ilja T. Feldstein

References

Andersen, G. J., & Enriquez, A. (2006). Aging and the detection of observer and moving object collisions. Psychology and Aging, 21(1), 74–85.

Anstis

(2003). Moving objects appear to slow down at low contrasts. Neural Networks, 16(5/6), 933–938.

Banister

Blackburn

J. M.

(1931). An eye factor affecting proficiency at ball games. British Journal of Psychology: General Section, 21(4), 382–384.

Barton, B. K., & Schwebel, D. C. (2007). The roles of age, gender, inhibitory control, and parental supervision in children's pedestrian safety. Journal of Pediatric Psychology, 32(5), 517–526.

Billington

Wilkie

R. M.

Field

D. T.

Wann

J. P.

(2010). Neural processing of imminent collision in humans. Proceedings of the Royal Society B: Biological Sciences, 278(1711), 1476–1481.

Blakemore

M. R.

Snowden

R. J.

(1999). The effect of contrast upon perceived speed: A general phenomenon? Perception, 28(1), 33–48.

Blakemore

M. R.

Snowden

R. J.

(2000). Textured backgrounds alter perceived speed. Vision Research, 40(6), 629–638.

Braunstein

M. L.

(1976). Depth perception through motion. New York, NY: Academic Press.

Bootsma

R. J.

Oudejans

R. R.

(1993). Visual information about time-to-collision between two objects. Journal of Experimental Psychology: Human Perception and Performance, 19(5), 1041–1052.

10.

Boring

E. G.

(1942). Sensation and perception in the history of experimental psychology. New York, NY: Apple-Century-Crofts.

11.

Caird

J. K.

Hancock

P. A.

(1994). The perception of arrival time for different oncoming vehicles at an intersection. Ecological Psychology, 6(2), 83–109.

12.

Carel

W. L.

(1961). Visual factors in the contact analog (Technical Report 263 820). Ithaca, NY: General Electric at Cornell University Industry Research Park.

13.

Cavallo

Laurent

(1988). Visual information and skill level in time-to-collision estimation. Perception, 17(5), 623–632.

14.

Cavanagh

Leclerc

Y. G.

(1989). Shape from shadows. Journal of Experimental Psychology: Human Perception and Performance, 15(1), 3–27.

15.

Connelly

M. L.

Conaglen

H. M.

Parsonson

B. S.

Isler

R. B.

(1998). Child pedestrians’ crossing gap thresholds. Accident Analysis & Prevention, 30(4), 443–453.

16.

Coull

J. T.

Vidal

Goulon

Nazarian

Craig

(2008). Using time-to-contact information to assess potential collision modulates both visual and temporal prediction networks. Frontiers in Human Neuroscience, 2, Article 10.

17.

Cutting

J. E.

Millard

R. T.

(1984). Three gradients and the perception of flat and curved surfaces. Journal of Experimental Psychology: General, 113(2), 198–216.

18.

Cutting

J. E.

Vishton

P. M.

(1995). Perceiving layout and knowing distances: The integration, relative potency, and contextual use of different information about depth. In Epstein

Rogers

(Eds.), Perception of space and motion (pp. 69–117). San Diego, CA: Academic Press.

19.

de Bellis

Schulte-Mecklenbeck

Brucks

Herrmann

Hertwig

(2018). Blind haste: As light decreases, speeding increases. PLoS One, 13(1), e0188951.

20.

DeLucia

P. R.

(1991). Pictorial and motion-based information for depth perception. Journal of Experimental Psychology: Human Perception and Performance, 17(3), 738–748.

21.

DeLucia

P. R.

(1999). Size-arrival effects: The potential roles of conflicts between monocular and binocular time-to-contact information, and of computer aliasing. Perception & Psychophysics, 61(6), 1168–1177.

22.

DeLucia

P. R.

(2004). Multiple sources of information influence time-to-contact judgments: Do heuristics accommodate limits in sensory and cognitive processes? In Hecht

Savelsbergh

G. J. P.

(Eds.), Advances in psychology, vol. 135: Time-to-contact (pp. 243–286). Amsterdam, the Netherlands: Elsevier.

23.

DeLucia

P. R.

(2005). Does binocular disparity or familiar size information override effects of relative size on judgements of time to contact? The Quarterly Journal of Experimental Psychology: Section A, 58(5), 865–886.

24.

DeLucia

P. R.

Kaiser

Bush

Meyer

Sweet

(2003). Information integration in judgements of time to contact. The Quarterly Journal of Experimental Psychology: Section A, 56(7), 1165–1189.

25.

DeLucia

P. R.

Preddy

Oberfeld

(2016). Audiovisual integration of time-to-contact information for approaching objects. Multisensory Research, 29(4/5), 365–395.

26.

DeLucia

P. R.

Warren

(1994). Pictorial and motion-based depth information during active control of self-motion: Size-arrival effects on collision avoidance. Journal of Experimental Psychology: Human Perception and Performance, 20(4), 783–798.

27.

Demetre

J. D.

Lee

D. N.

Grieve

Pitcairn

T. K.

Ampofo-Boateng

Thomson

J. A.

(1993). Young children’s learning on road-crossing simulations. British Journal of Educational Psychology, 63(2), 349–359.

28.

Demetre, J. D., Lee, D. N., Pitcairn, T. K., Grieve, R., Thomson, J. A., & Ampofo-Boateng, K. (1992). Errors in young children's decisions about traffic gaps: Experiments with roadside simulations. British Journal of Psychology, 83(2), 189–202.

29.

Dommes

Cavallo

(2011). The role of perceptual, cognitive, and motor abilities in street-crossing decisions of young and older pedestrians. Ophthalmic and Physiological Optics, 31(3), 292–301.

30.

Dommes

Cavallo

(2012). Can simulator-based training improve street-crossing safety for elderly pedestrians? Transportation Research Part F: Traffic Psychology and Behaviour, 15(2), 206–218.

31.

Dommes

Cavallo

Dubuisson

J. B.

Tournier

Vienne

(2014). Crossing a two-way street: Comparison of young and old pedestrians. Journal of Safety Research, 50, 27–34.

32.

Dommes

Cavallo

Oxley

(2013). Functional declines as predictors of risky street-crossing decisions in older pedestrians. Accident Analysis & Prevention, 59, 135–143.

33.

Dommes

Cavallo

Vienne

Aillerie

(2012). Age-related differences in street-crossing safety before and after training of older pedestrians. Accident Analysis & Prevention, 44(1), 42–47.

34.

Dommes

Le Lay

Vienne

Dang

N. T.

Beaudoin

A. P.

M. C.

(2015). Towards an explanation of age-related difficulties in crossing a two-way street. Accident Analysis & Prevention, 85, 229–238.

35.

Ferris

S. H.

(1972). Motion parallax and absolute distance. Journal of Experimental Psychology, 95(2), 258–263.

36.

Field

(2013). Discovering statistics using IBM SPSS statistics (4th ed.). London, England: Sage Publications.

37.

Field

D. T.

Wann

J. P.

(2005). Perceiving time to collision activates the sensorimotor cortex. Current Biology, 15(5), 453–458.

38.

Fisher

S. K.

Ciuffreda

K. J.

(1988). Accommodation and apparent distance. Perception, 17(5), 609–621.

39.

Foley

J. M.

(1991). Stereoscopic distance perception. In Ellis

S. R.

Kaiser

M. K.

Grunwald

A. J.

(Eds.), Pictorial communication in virtual and real environments (pp. 559–566). London, England: Taylor & Francis.

40.

Gegenfurtner

K. R.

Mayser

Sharpe

L. T.

(1999). Seeing movement in the dark. Nature, 398(6727), 475–476.

41.

Geri

G. A.

Gray

Grutzmacher

(2010). Simulating time-to-contact when both target and observer are in motion. Displays, 31(2), 59–66.

42.

Gibson

J. J.

(1950). The perception of the visual world. Boston, MA: Houghton Mifflin.

43.

González

E. G.

Allison

R. S.

Ono

Vinnikov

(2010). Cue conflict between disparity change and looming in the perception of motion in depth. Vision Research, 50(2), 136–143.

44.

Gray

Regan

(1998). Accuracy of estimating time to collision using binocular and monocular information. Vision Research, 38(4), 499–512.

45.

Green

D. M.

Swets

J. A.

(1966). Signal detection theory and psychophysics. New York, NY: John Wiley & Sons.

46.

Groeger

J. A.

Cavallo

(1991). Judgements of time-to-collision and time-to-coincidence. In Gale

A. G.

Brown

I. D.

Haslegrave

C. M.

Moorhead

Taylor

(Eds.), Vision in vehicles III (pp. 27–34). Amsterdam, the Netherlands: Elsevier.

47.

Hancock

P. A.

Manser

M. P.

(1997). Time-to-contact: More than tau alone. Ecological Psychology, 9(4), 265–297.

48.

Hancock

P. A.

Manser

M. P.

(1998). Time-to-contact. In Feyer

A.-M.

Williamson

(Eds.), Occupational injury: Risk, prevention, and intervention (pp. 44–59). London, England: Taylor & Francis.

49.

Hills

B. L.

(1980). Vision, visibility, and perception in driving. Perception, 9(2), 183–216.

50.

Horswill

M. S.

Helman

Ardiles

Wann

J. P.

(2005). Motorcycle accident risk could be inflated by a time to arrival illusion. Optometry and Vision Science, 82(8), 740–746.

51.

Kadali

B. R.

Vedagiri

(2012). Pedestrians’ gap acceptance behavior at mid block location. International Journal of Engineering and Technology, 4(2), 158–161.

52.

Kaiser

M. K.

Mowafy

(1993). Optical specification of time-to-passage: Observers’ sensitivity to global tau. Journal of Experimental Psychology: Human Perception and Performance, 19(5), 1028–1040.

53.

Kaplan

G. A.

(1969). Kinetic disruption of optical texture: The perception of depth at an edge. Perception & Psychophysics, 6(4), 193–198.

54.

Kappé

Korteling

J. E.

(1995). Perception of time to contact in driving simulators (Report TNO-TM 1995 A-53). Soesterberg, the Netherlands: TNO Human Factors Research Institute.

55.

Khuu

S. K.

Lee

T. C.

Hayes

(2010). Object speed derived from the integration of motion in the image plane and motion-in-depth signaled by stereomotion and looming. Vision Research, 50(9), 904–913.

56.

Kirkham

R. W.

Landauer

A. A.

(1985). Sex differences in the distribution of traffic law enforcement. Accident Analysis & Prevention, 17(3), 211–215.

57.

Konečni

Ebbeson

E. B.

Konečni

D. K.

(1976). Decision processes and risk taking in traffic: Driver response to the onset of yellow light. Journal of Applied Psychology, 61(3), 359–367.

58.

Kubovy

(1988). The psychology of perspective and renaissance art. Cambridge, England: Cambridge University Press.

59.

Lappin

J. S.

(2014). What is binocular disparity? Frontiers in Psychology, 5, Article 870.

60.

Lee

D. N.

(1976). A theory of visual control of braking based on information about time-to-collision. Perception, 5(4), 437–459.

61.

Lee, D. N., Young, D. S., & McLaughlin, C. M. (1984). A roadside simulation of road crossing for children. Ergonomics, 27(12), 1271–1281.

62.

F. X.

Laurent

(1995). Occlusion rate of ball texture as a source of velocity information. Perceptual and Motor Skills, 81(3), 871–880.

63.

Lobjois

Cavallo

(2007). Age-related differences in street-crossing decisions: The effects of vehicle speed and time constraints on gap selection in an estimation task. Accident Analysis & Prevention, 39(5), 934–943.

64.

Lobjois

Cavallo

(2009). The effects of aging on street-crossing behavior: From estimation to actual crossing. Accident Analysis & Prevention, 41(2), 259–267.

65.

López-Moliner

Brenner

Smeets

J. B.

(2007). Effects of texture and shape on perceived time to passage: Knowing “what” influences judging “when”. Perception & Psychophysics, 69(6), 887–894.

66.

Lugtigheid

A. J.

Welchman

A. E.

(2011). Evaluating methods to measure time-to-contact. Vision Research, 51(20), 2234–2241.

67.

Manser

M. P.

Hancock

P. A.

(1996). Influence of approach angle on estimates of time-to-contact. Ecological Psychology, 8(1), 71–99.

68.

Marr

(2010). Vision: A computational investigation into the human representation and processing of visual information. Cambridge, MA: MIT Press.

69.

Mathieu

Bootsma

R. J.

Berthelon

Montagne

(2017). Judging arrival times of incoming traffic vehicles is not a prerequisite for safely crossing an intersection: Differential effects of vehicle size and type in passive judgment and active driving tasks. Acta Psychologica, 173, 1–12.

70.

Matthews, M. L. (1986). Aging and the perception of driving risk and ability. In Proceedings of the Human Factors Society Annual Meeting, 30(12), 1159–1163.

71.

McGuckin

Fucci

(2018). Summary of travel trends: 2017 National household travel survey (Report FHWA-PL-18-019). Washington, DC: Federal Highway Administration, U.S. Department of Transportation.

72.

McKee

S. P.

(1981). A local mechanism for differential velocity detection. Vision Research, 21(4), 491–500.

73.

McLeod

R. W.

Ross

H. E.

(1983). Optic-flow and cognitive factors in time-to-collision estimates. Perception, 12(4), 417–423.

74.

Molyneux

(1690). Dipotrica nova. A treatise of dioptricks, in two parts. London, England: Benjamin Tooke.

75.

Montgomery

Kusano

K. D.

Gabler

H. C.

(2014). Age and gender differences in time to collision at braking from the 100-car naturalistic driving study. Traffic Injury Prevention, 15(Supp. 1), 15–20.

76.

Nagata

(1989). How to reinforce perception of depth in single two-dimensional pictures. In Ellis

S. R.

Kaiser

M. K.

Grunwald

A. J.

(Eds.), Spatial displays and spatial instruments (Art. 20, pp. 1–18). Moffett Field, CA: National Aeronautics and Space Administration.

77.

Okoshi

(1976). Three-dimensional imaging techniques. New York, NY: Academic Press.

78.

O'Neal

E. E.

Jiang

Franzen

L. J.

Rahimian

Yon

J. P.

Kearney

J. K.

Plumert

J. M.

(2018). Changes in perception–action tuning over long time scales: How children and adults perceive and act on dynamic affordances when crossing roads. Journal of Experimental Psychology: Human Perception and Performance, 44(1), 18–26.

79.

Oxley, J. A., Ihsen, E., Fildes, B. N., Charlton, J. L., & Day, R. H. (2005). Crossing roads safely: an experimental study of age differences in gap selection by pedestrians. Accident Analysis & Prevention, 37(5), 962–971.

80.

Palmisano

Gillam

Govan

D. G.

Allison

R. S.

Harris

J. M.

(2010). Stereoscopic perception of real depths at large distances. Journal of Vision, 10(6), Article 19.

81.

Parsonson

B. S.

Isler

R. B.

Hansson

G. J.

(1996). Driver behaviour at rural T-intersections (Report No. 56). Hamilton, New Zealand: University of Waikato.

82.

Petzoldt

(2014). On the relationship between pedestrian gap acceptance and time to arrival estimates. Accident Analysis & Prevention, 72, 127–133.

83.

Plumert

J. M.

Kearney

J. K.

Cremer

J. F.

(2004). Children’s perception of gap affordances: Bicycling across traffic-filled intersections in an immersive virtual environment. Child Development, 75(4), 1243–1253.

84.

Polus

Hocherman

Efrat

(1988). Evaluation of the accident rates of male and female drivers. Transportation Research Record, 1172, 42–46.

85.

Rand

K. M.

Tarampi

M. R.

Creem-Regehr

S. H.

Thompson

W. B.

(2011). The importance of a visual horizon for distance judgments under severely degraded vision. Perception, 40(2), 143–154.

86.

Recarte

M. A.

Conchillo

Nunes

L. M.

(2005). Estimation of arrival time in vehicle and video. Psicothema, 17(1), 112–117.

87.

Regan

Hamstra

S. J.

(1993). Dissociation of discrimination thresholds for time to contact and for rate of angular expansion. Vision Research, 33(4), 447–462.

88.

Regan

Vincent

(1995). Visual processing of looming and time to contact throughout the visual field. Vision Research, 35(13), 1845–1857.

89.

Renner

R. S.

Velichkovsky

B. M.

Helmert

J. R.

(2013). The perception of egocentric distances in virtual environments: A review. ACM Computing Surveys, 46(2), Article 23.

90.

Satava, R. M., & Jones, S. (2003) Medical applications in virtual reality. In L. J. Hettinger, & M. W. Haas (Eds.), Virtual and Adaptive Environments: Applications, Implications, and Human Performance Issues (pp. 325–343). Mahwah, NJ: Lawrence Erlbaum Associates.

91.

Scarfe

Glennerster

(2015). Using high-fidelity virtual reality to study perception in freely moving observers. Journal of Vision, 15(9), Article 3.

92.

Schiff

(1980). Perception: An applied approach. Boston, MA: Houghton Mifflin.

93.

Schiff

Detwiler

M. L.

(1979). Information used in judging impending collision. Perception, 8(6), 647–658.

94.

Schiff

Oldak

(1990). Accuracy of judging time to arrival: Effects of modality, trajectory, and gender. Journal of Experimental Psychology: Human Perception and Performance, 16(2), 303–316.

95.

Schiff

Oldak

Shah

(1992). Aging persons’ estimates of vehicular motion. Psychology and Aging, 7(4), 518–525.

96.

Schwebel

D. C.

Gaines

Severson

(2008). Validation of virtual reality as a tool to understand and prevent child pedestrian injury. Accident Analysis & Prevention, 40(4), 1394–1400.

97.

Sedgwick

H. A.

(1983). Environment-centered representation of spatial layout: Available visual information from texture and perspective. In Beck

Hope

Rosenfeld

(Eds.), Human and machine vision (pp. 425–458). New York, NY: Academic Press.

98.

Seward

A. E.

Ashmead

D. H.

Bodenheimer

(2007). Using virtual environments to assess time-to-contact judgments from pedestrian viewpoints. ACM Transactions on Applied Perception (TAP), 4(3), Article 18.

99.

Sidaway

Fairweather

Sekiya

Mcnitt-Gray

(1996). Time-to-collision estimation in a simulated driving task. Human Factors, 38(1), 101–113.

100.

Snowden

R. J.

Stimpson

Ruddle

R. A.

(1998). Speed perception fogs up as visibility drops. Nature, 392(6675), 450.

101.

Sousa

Brenner

Smeets

J. B.

(2011). Judging an unfamiliar object’s distance from its retinal image size. Journal of Vision, 11(9), Article 10.

102.

Sousa

Smeets

J. B.

Brenner

(2012). Does size matter? Perception, 41(12), 1532–1534.

103.

Staplin

(1995). Simulator and field measures of driver age differences in left-tum gap judgments. Transportation Research Record, 1485, 49–55.

104.

Stevens

K. A.

(1981). The information content of texture gradients. Biological Cybernetics, 42(2), 95–105.

105.

Stevens

S. S.

(1946). On the theory of scales of measurement. Science, 103(2684), 677–680.

106.

Stewart

Cudworth

C. J.

Lishman

J. R.

(1993). Misperception of time-to-collision by drivers in pedestrian accidents. Perception, 22(10), 1227–1244.

107.

Stone

L. S.

Thompson

(1992). Human speed perception is contrast dependent. Vision Research, 32(8), 1535–1549.

108.

Takeuchi

De Valois

K. K.

(2000). Velocity discrimination in scotopic vision. Vision Research, 40(15), 2011–2024.

109.

Tharanathan, A., & DeLucia, P. R. (2006). Time-to-Collision Judgments of Constant and Non-Constant Velocities: Implications for Rear-End Collisions. In Proceedings of the Human Factors and Ergonomics Society Annual Meeting, 50(22), 2463–2467.

110.

Thompson

(1982). Perceived rate of movement depends on contrast. Vision Research, 22(3), 377–380.

111.

Thomson

J. A.

(1983). Is continuous visual monitoring necessary in visually guided locomotion? Journal of Experimental Psychology: Human Perception and Performance, 9(3), 427–443.

112.

Thomson

J. A.

Tolmie

A. K.

Foot

H. C.

Whelan

K. M.

Sarvary

Morrison

(2005). Influence of virtual reality training on the roadside crossing judgments of child pedestrians. Journal of Experimental Psychology: Applied, 11(3), 175–186.

113.

Todd

J. T.

(1981). Visual information about moving objects. Journal of Experimental Psychology: Human Perception and Performance, 7(4), 795–810.

114.

Tresilian

J. R.

(1991). Empirical and theoretical issues in the perception of time to contact. Journal of Experimental Psychology: Human Perception and Performance, 17(3), 865–876.

115.

Tresilian

J. R.

(1995). Perceptual and cognitive processes in time-to-contact estimation: Analysis of prediction-motion and relative judgment tasks. Perception & Psychophysics, 57(2), 231–245.

116.

van der Kamp

Savelsbergh

Smeets

(1997). Multiple information sources in interceptive timing. Human Movement Science, 16(6), 787–821.

117.

Vincent

Regan

(1997). Judging the time to collision with a simulated textured object: Effect of mismatching rate of expansion of object size and of texture element size. Perception & Psychophysics, 59(1), 32–36.

118.

von Hofsten

(1976). The role of convergence in visual space perception. Vision Research, 16(2), 193–198.

119.

Voyer

Bryden

M. P.

(1995). Magnitude of sex differences in spatial abilities: A meta-analysis and consideration of critical variables. Psychological Bulletin, 117(2), 250–270.

120.

Wallach

O'Connell

D. N.

(1953). The kinetic depth effect. Journal of Experimental Psychology, 45(4), 205–217.

121.

Warren

W. H.

Hannon

D. J.

(1988). Direction of self-motion is perceived from optical flow. Nature, 336(6195), 162–163.

122.

Watson

J. S.

Banks

M. S.

von Hofsten

Royden

C. S.

(1992). Gravity as a monocular cue for perception of absolute distance and/or absolute size. Perception, 21(1), 69–76.

123.

Yan

J. J.

Lorv

Sun

H. J.

(2011). Visual processing of the impending collision of a looming object: Time to collision revisited. Journal of Vision, 11(12), Article 7.

124.

Young, D. S., & Lee, D. N. (1987). Training children in road crossing skills using a roadside simulation. Accident Analysis & Prevention, 19(5), 327–341.