Potential Effectiveness of Bicycle-Automatic Emergency Braking using the Washtenaw Area Transportation Study Data Set

Abstract

In the United States, fatalities from vehicle–bicycle crashes have been increasing since 2010. A total of 857 cyclists were struck and killed in 2018 which is an increase from 623 fatalities in 2010. One promising countermeasure is Automatic Emergency Braking (AEB), which can help prevent and/or mitigate many vehicle–bicycle crashes. AEB is a vehicle-based system that can detect and mitigate an impending crash. The goal of this study was to elucidate U.S. vehicle–bicycle crashes and examine related factors to estimate AEB effectiveness. This study used a unique in-depth vehicle–bicycle crash study dataset collected under the collaboration of the Washtenaw Area Transportation Study (WATS) and the Toyota Collaborative Research Center conducted in southeast Michigan from 2011 to 2013. The WATS database provides in-depth investigations of vehicle–bicycle crashes in the United States. The characteristics of the WATS vehicle–bicycle crashes were validated against the Fatality Analysis Reporting System and the General Estimate System. The WATS database cases were examined to estimate the potential effectiveness of AEB to prevent or mitigate vehicle–bicycle collisions. In 60% of the WATS cases, cyclists were in the road for more than 1 s before impact. Assuming that a hypothetical AEB system requires a minimum of 1 s for detection and brake activation, these collisions would potentially be avoided or mitigated. However, for the remaining cases with less than 1 s of time to react (40% of cases), that AEB system would be challenged to avoid or mitigate the collision.

In the United States, fatalities from vehicle–bicycle crashes have been increasing since 2010. A total of 857 cyclists were struck and killed in 2018 which is an increase from 623 fatalities in 2010 ( 1 , 2 ). Cyclists consistently account for about 2% of all U.S. traffic fatalities. From 1997 to 2013 the annual cost of non-fatal bicycle injuries increased by an average of $800 million a year (in 2010 dollars) with 2013 costs reaching an estimated $22.3 billion in 2010 dollars ( 3 ). One promising active safety countermeasure is Automatic Emergency Braking (AEB). AEB is a vehicle-based system that uses a combination of forward facing sensors, commonly radar and cameras, to detect an impending crash with a cyclist and automatically apply the brakes to avoid or mitigate the crash ( 4 ).

Using European data, Rosén ( 5 ) examined the potential effectiveness of a range of AEB system designs to mitigate 607 vehicle–bicycle crashes extracted from the German In-Depth Accident Study (GIDAS) database. The four main factors that affected AEB effectiveness were found to be the vehicle speed, lighting conditions, time to collision, and deceleration ability. The study found that when the vehicle’s braking system was applied at maximum capacity, 84% of the fatalities could have been prevented. When the vehicle’s braking system was applied at minimal capacity, 19% of fatalities could have been prevented ( 5 ). These estimates in fatality reduction with respect to braking demonstrate how applying the brakes at maximum capacity can better reduce the number of fatalities. Fredriksson and Rosén ( 6 ) used GIDAS to find a relationship between body region injury and impact location on the vehicle. Both of these studies relied on German crash data and therefore may not be directly applicable to U.S. crashes as the U.S. vehicle fleet and infrastructure may differ from Germany’s. For example, many U.S. states allow “right turn on red” which is not permitted in Germany ( 7 ).

Unfortunately, little in-depth vehicle–bicycle crash data is available to study vehicle–bicycle crashes in the United States. In lieu of in-depth crash data, U.S. vehicle–bicycle studies have examined naturalistic and government curated national databases. Sherony et al. ( 8 ) used naturalistic data from the Transportation Active Safety Institute (TASI) 110-Car study to examine cyclist motion during vehicle–bicycle interactions to aid in the design of AEB cyclist detection systems. Haus and Gabler examined U.S. vehicle–bicycle near crashes in the second Strategic Highway Research Program (SHRP 2), a naturalistic driving study, and found that in about half of the cases the cyclist was visible for at least 1 s which is promising for AEB as a mitigation strategy ( 9 ). An analysis of U.S. national databases from 2014 to 2017 found that 71% of vehicle–bicycle crashes and 82% of fatal vehicle–bicycle crashes were frontal crashes and therefore potentially mitigatable by AEB ( 10 ).

The goal of this study was to elucidate U.S. vehicle–bicycle crashes and examine factors related to AEB effectiveness using in-depth vehicle–bicycle crash data.

Methods

This study used a new data set collected under the Washtenaw Area Transportation Study (WATS) conducted in southeast Michigan. This data set was collected in collaboration with Toyota Collaborative Safety Research Center (CSRC). First, WATS was compared with the Fatality Analysis Reporting System (FARS) and the National Automotive Sampling System (NASS) – General Estimate System (GES) to compare and contrast WATS findings with national observations. Then, the WATS database cases were examined to determine potential AEB effectiveness.

Data Sources

The data for this study was compiled from three data sources (shown in Table 1): the WATS, the FARS, and the NASS-GES.

Table 1.

Overview of Data Sources

Data set	WATS	FARS	GES
Injury severity	All injury levels	Fatal	All injury levels
Years	2011–2013	2011–2015	2011–2015
Total cases (unweighted)	100	3,554	8,353
Weighted	No	No	Yes
Nationally representative	No	Yes	Yes
Sample	Sample of crashes in southeast Michigan	Census of U.S. fatal crashes	All U.S. police-reported crashes
Scene diagram	Yes	No	No
Imputed variables	No	No	Yes

Note: WATS = Washtenaw Area Transportation Study; FARS = Fatality Analysis Reporting System; GES = General Estimate System.

The WATS data set, collected in collaboration with the Toyota Collaborative Safety Research Center, contains a sample of vehicle–bicycle crashes that occurred in southeast Michigan from 2011 to 2013. The data set contains both vehicle–bicycle crashes and vehicle-pedestrian crashes. Only vehicle–bicycle crashes were analyzed in this study. Characteristics of the drivers, cyclists, environments, events, and detection times were examined to determine factors which could influence the type and severity of vehicle–bicycle crashes. WATS cases were excluded from the data set if they were missing any data elements. The final data set used for this study included 100 vehicle–bicycle crashes. A summary of the WATS data is included in Table 1 and detailed data composition tables are provided in the Supplemental Appendix.

FARS is a census of all U.S. vehicle-related fatalities that occurred on public roads. This study examined data from 2011 to 2015, which contained records of 3,554 cyclist fatalities. FARS includes information on environmental conditions, the persons involved, and crash causation factors. FARS does not however include detailed reconstructions or crash scene diagrams.

GES is a probability sample of all police-reported crashes that occurred on U.S. public roads. Each GES case includes a sampling weight which can be applied to individual cases to estimate national incidence. Unless otherwise stated, the numbers reported in this study are the weighted estimates. GES contains crashes of all injury severities ranging from property damage to fatal injury. This study examined data from 2011 to 2015, which contained 8,353 cases corresponding to 267,440 crashes. GES contains similar data elements to FARS, and, like FARS, does not include crash scene diagrams.

Calculate Earliest Detection Opportunity

The effectiveness of AEB depends on the ability of the sensors to detect the cyclist, but this information is not available in U.S. databases. We used the WATS scene diagrams and associated crash event records to estimate time to collision, that is, how long the cyclist was in the road before the crash.

The estimated time to collision, referred to here as the Earliest Detection Opportunity (EDO), was calculated by dividing the distance the cyclist traveled in the road by the cyclist riding velocity as shown in Equation 1:

\begin{matrix} EDO = \frac{distance}{velocit y_{bicyclist}} \end{matrix}

(1)

For a cyclist crossing scenario, the distance the cyclist traveled in the road before impact was measured from the crash scene diagrams. As shown in Figure 1, each crash scene diagram was uploaded into AutoCAD and scaled to proper dimensions. The cyclist’s path was then measured from the edge of the road to the impact point.

Figure 1.

Sample crash scene diagram and AutoCAD measurement. Green bracket indicates the measured distance the cyclist traveled in the road.

For cyclists riding along the road, relative speed was used instead of the cyclist’s speed. The cyclist was assumed to be visible for 40 m. Although current radar systems can detect pedestrians at ranges upwards of 90 m in ideal conditions ( 11 ). A detection range of 40 m was chosen because camera-based detection and non-ideal conditions can reduce pedestrian detection ranges ( 12 , 13 ). The relative speed was the difference between the cyclist’s estimated speed and the vehicle’s estimated speed as shown in Equation 2. In our study, the posted speed limit was used as a surrogate for the vehicle speed as the actual travel speed of the vehicle was not reported in the WATS data set.

\begin{matrix} EDO = \frac{distance}{velocit y_{vehicle} - velocit y_{bicyclist}} \end{matrix}

(2)

The speeds of the cyclists were assumed to depend on age, gender, and crossing behavior. Age and gender were available directly from the WATS database. Cyclist crossing behavior was determined based on the crash scene diagram and associated event information. Cyclist crossing behavior was grouped into three categories: riding along the road, riding through an intersection, or turning left/right (Figure 2). Riding through was defined as the cyclist traveling through an intersection. Riding along was defined as the cyclist traveling parallel to the vehicle on the road.

Figure 2.

Diagrams: (left) ride through; (center) turning left/right; and (right) ride along.

Thompson et al. ( 14 ) reported the cyclist’s speed as a function of age and gender. The findings were based on radar-gun measurement of the speed of 152 children and adults riding bicycles on a closed road. Ling and Wu ( 15 ) calculated the speeds for cyclists turning left through analysis of video surveillance of 531 cyclists at intersections. Sherony et al. ( 16 ) determined the speeds of 895 cyclists riding through/along and at certain parts through the road from an analyzing the TASI 110. The TASI 110 car study was a one year naturalistic driving collection conducted in the greater Indianapolis area ( 17 ). For this study, the 50th percentile speed was used for each characteristic of the cyclist since it represents the average rider. The cyclist values used are shown in Tables 2 and 3.

Table 2.

Average Cyclist Speeds for Riding Along the Road and Turning ( 14 , 15 )

	Sample size	Male (m/s)	Female (m/s)	Unknown gender (m/s)
≤13 Years old	107	4.11	3.84	3.97
≥14 Years old	45	4.69	4.04	4.33
Unknown age	152	4.29	3.93	5.6
Turning left	531	3.58	3.73	3.66

Table 3.

Average Cyclist Speeds for Riding Through an Intersection ( 18 )

Sample size: 280	Starting stationary (m/s)	Constant speed (m/s)
25% Across road	3.35	4
50% Across road	3.77	5.2
75% Across road	3.85	6.3
Riding through	na	5.2

Note: na = not applicable.

Results

The WATS data set was compared with FARS and GES data on driver/cyclist, environment, and event characteristics. As anticipated, the WATS data set more closely resembled GES than FARS as both WATS and GES contained crashes with all injury severities. In contrast, FARS only contained fatal collisions. A chi-squared analysis was conducted across driver, cyclist, environmental, and event characteristics between WATS and GES. The chi-squared analysis revealed no significant difference between driver gender, driver alcohol involvement, cyclist alcohol involvement, cyclist injury severity, road surface conditions, lighting, relation to intersection, traffic control vehicle first impact location, and vehicle pre-event movement distributions. There was a significant difference between WATS and GES distributions of weather conditions and cyclist gender. WATS had a higher percentage of female riders and cloudy weather conditions. Further discussion of these distributions can be found in the discussion section and the full analysis can be found in the Supplemental Appendix.

Turning crashes may present a challenge for AEB systems as the vehicle-mounted sensors may have limited field of view to detect the cyclists. In turning cases, the cyclist is often not in view until just before impact. As shown in Figure 3, most drivers were going straight or turning before the crash which likely resulted in front or side first impacts across all three data sets. The WATS data were similar to GES with about 30% to 40% of cars traveling straight and about 25% turning right. In FARS, however, more than three-fourths (78%) of fatal crashes occurred when the vehicle was traveling straight and very few occurred when the motorist was turning left or right.

Figure 3.

Distribution of vehicle pre-event movements before colliding with a cyclist.

In addition to detecting the cyclist, bicycle AEB systems must be able to predict the cyclist’s movement. This presents a challenge given the multitude of ways cyclists and vehicles interact and the possibility of that cyclist being in the road or on pedestrian pathways. Most vehicle–bicycle collisions in WATS occurred at intersections (76 of the 100 crashes). As shown in Figure 4, crashes that involved the cyclist riding along the road only resulted in 20 collisions, which was approximately one-fourth the number of intersection crashes. There were only three cases in which the cyclist was turning: two left-turning cases and one right-turning case. Turning left may put the cyclist more at risk because the cyclist had to cross another through lane, but this cannot be confirmed with this small sample of turning cases. The most common vehicle pre-crash movement was turning right, closely followed by turning left. Ride through and ride along vehicle travel paths indicate the vehicle was traveling straight through an intersection or straight on a non-junction, respectively.

Figure 4.

Percentage of crash cases for varying vehicle and cycling traveling paths in Washtenaw Area Transportation Study.

Once the cyclist is detected, the effectiveness of the AEB system largely depends on the time the system has to react. The longer a cyclist is visible to an AEB system, the longer the system has to initiate braking and avoid the collision. As shown in Figure 5, the cyclist ride along and ride through cases have similar distributions of earliest detection time. For ride along cases, the cyclist was visible for at least 0.5 s, 1 s, and 2 s for 25%, 45%, and 70% of the cases, respectively. For ride through cases, cyclists were visible for at least 0.5 s, 1 s, and about 2 s for 23%, 39%, and 69% of the cases, respectively. While AEB system parameters are highly proprietary and differ by manufacturer, 1 s is a commonly assumed threshold above which an AEB system could mitigate or avoid a crash. If an AEB system only needs 1 s to mitigate a crash, then about 60% of the crashes in the WATS database could potentially be mitigated.

Figure 5.

Cumulative distribution of earliest detection time for ride along and ride through cases in Washtenaw Area Transportation Study.

Discussion

This study examined vehicle–bicycle crashes investigated by WATS in southeast Michigan with support from Toyota CSRC. The goal of this study was to elucidate U.S. vehicle–bicycle crashes and examine factors that could be detrimental to AEB effectiveness. The data from WATS was able to provide many details about vehicle–bicycle crashes that are not available in national databases and could be very useful in research and further development of AEB.

Although WATS data collection was limited to southeast Michigan, WATS was broadly similar to the nationally representative GES. The two exceptions were weather and cyclist gender which had statistically significant differences between WATS and GES. While weather conditions, for example, rain, could affect AEB performance, the distribution of road surface conditions, for example, wet versus dry pavement, was not significantly different. The difference may reflect actual differences in weather at the time of collision, but may also reflect how quickly investigators were able to visit the scene after the collision. WATS and GES also had significantly different distributions of cyclist gender, although cyclist gender could affect cyclist speed, it would likely have little effect on crash causation factors or AEB detection capabilities.

The WATS database provides a unique data set of in-depth investigation of vehicle–bicycle crashes in the United States which allows computation of important parameters such as EDO and distance traveled in the road. This study was limited by the number of cases (100) in the WATS data set. Another limitation in the study was the lack of available travel speed data in the WATS data set. For cyclists riding along the road, our study used posted speed limits as a surrogate for the driver travel speed. This assumption may affect the earliest detection calculation. For all cases, cyclist speed was not known and therefore had to be estimated based on literature values. There were two cases where the cyclist’s speed and earliest detection time could not be calculated and were omitted from our calculations.

This work presents variables that can contribute to the potential effectiveness of AEB in reducing or mitigating vehicle–bicycle crashes. To estimate effectiveness further research needs to examine the time requirements for braking, the effects of pavement conditions on maximum possible braking, and the effects of obstructions. WATS does not record the impact speed or travel speed of the vehicle which make it difficult to simulate the crashes accurately. While WATS is more in-depth than publicly available vehicle–bicycle crash databases in the United States, a national study that includes a larger sample size, crash reconstructions, and more detailed injury information would benefit AEB effectiveness estimations.

Conclusion

AEB is a promising active safety technology that may avoid or mitigate many vehicle-to-bicycle collisions. Cyclists were in the road for more than 1 s before impact in 60% of the WATS cases. Assuming that an AEB system requires a minimum of 1 s for detection and brake activation, these collisions would be potentially detectable by an AEB system with sufficient time to take evasive action. However, for the remaining cases with less than 1 s of time to react (40% of cases), an AEB system would be challenged to avoid or mitigate the collision. Collision type, such as straight crossing path or turning, could also affect AEB effectiveness.

Supplemental Material

sj-docx-1-trr-10.1177_03611981211001377 – Supplemental material for Potential Effectiveness of Bicycle-Automatic Emergency Braking using the Washtenaw Area Transportation Study Data Set

Supplemental material, sj-docx-1-trr-10.1177_03611981211001377 for Potential Effectiveness of Bicycle-Automatic Emergency Braking using the Washtenaw Area Transportation Study Data Set by Samantha H. Haus, Ryan M. Anderson, Rini Sherony and Hampton C. Gabler in Transportation Research Record

Footnotes

Acknowledgements

The authors would like to acknowledge the Toyota Collaborative Safety Research Center (CSRC) and Toyota Motor Corporation for funding and providing technical support for this study.

Author Contributions

The authors confirm contribution to the paper as follows: study conception and design: Hampton C. Gabler; collected additional variables from dataset: Ryan M. Anderson; analysis and interpretation of results: Ryan M. Anderson, and Samantha H. Haus; draft manuscript preparation: Samantha H. Haus, Ryan M. Anderson, Rini Sherony, and Hampton C. Gabler. All authors reviewed the results and approved the final version of the manuscript.

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: The authors received financial support from the Toyota Collaborative Safety Research Center (CSRC).

Supplemental Material

Supplemental material for this article is available online.

References

National Center for Statistics and Analysis. Traffic Safety Facts: Cyclists and Other Cyclists (2017 Data). U.S. Department of Transportation, Washington, D.C., June 2019.

National Center for Statistics and Analysis. 2018 Fatal Motor Vehicle Crashes: Overview (Traffic Safety Facts). National Highway Traffic Safety Administration, Washington, D.C., 2019.

Gaither

T. W.

Sanford

T. A.

Awad

M. A.

Osterberg

E. C.

Murphy

G. P.

Lawrence

B. A.

Miller

T. R.

Breyer

B. N.

Estimated Total Costs From Non-Fatal and Fatal Bicycle Crashes in the USA: 1997–2013. Injury Prevention: Journal of the International Society for Child and Adolescent Injury Prevention, Vol. 24, No. 2, 2018, pp. 135–141.

Schram

Williams

Ratingen

M. V.

Ryrberg

Sferco

, eds. Euro NCAP’s First Step to Assess Autonomous Emergency Braking (AEB) for Vulnerable Road Users. Proc., 24th International Technical Conference on the Enhanced Safety of Vehicles (ESV), Gothenburg, Sweden, 2015.

Rosén

, ed. Autonomous Emergency Braking for Vulnerable Road Users. Proc., IRCOBI Conference, 2013; Gothenburg, Sweden.

Fredriksson

Rosén

, eds. Priorities for Cyclist Protection in Car Impacts: A Real Life Study of Severe Injuries and Car Sources. Proc., IRCOBI Conference, 2012, Dublin, Ireland.

Road Traffic Regulations (Straßenverkehrs-Ordnung, StVO) Section 37.(2).1, 2016. https://germanlawarchive.iuscomp.org/?p=1290.

Sherony

Tian

Chien

Chen

Takahashi

Pedestrian/Cyclist Limb Motion Analysis from 110-Car TASI Video Data for Autonomous Emergency Braking Testing Surrogate Development. SAE International Journal of Transportation Safety, Vol. 4, No. 1, 2016, pp. 113–120.

Haus

S. H.

Gabler

H. C.

, eds. Characteristics of Vehicle–bicycle Crashes and Near Crashes Using Naturalistic Driving Data. Presented at Transportation Research Board 97th Annual Meeting, Washington, D.C., 2018.

10.

Haus

S. H.

Gabler

H. C.

, eds. The Potential for Active Safety Mitigation of US Vehicle–bicycle Crashes. In Future Active Safety Technology Towards Zero Traffic Accidents (FAST-Zero), Blacksburg, VA, 2019.

11.

System Reference Document (SRdoc); Transmission Characteristics; Technical Characteristics for Radiodetermination Equipment for Ground Based Vehicular Applications Within the Frequency Range 77 GHz to 81 GHz, 2020. Contract No.: ETSI TR 103 593. https://www.etsi.org/deliver/etsi_tr/103500_103599/103593/01.01.01_60/tr_103593v010101p.pdf.

12.

Belyaev

A. A.

Suanov

T. A.

Frolov

I. O.

Trots

D. O.

, eds. The Range of Pedestrian Detection with Automotive Radar. Proc., 2019 Radiation and Scattering of Electromagnetic Waves (RSEMW), June24–28, 2019.

13.

Mohammed

A. S.

Amamou

Ayevide

F. K.

Kelouwani

Agbossou

Zioui

The Perception System of Intelligent Ground Vehicles in All Weather Conditions: A Systematic Literature Review. Sensors (Basel, Switzerland), Vol. 20, No. 22, 2020, pp. 8–22. https://pubmed.ncbi.nlm.nih.gov/33203155/.

14.

Thompson

D. C.

Rebolledo

Thompson

R. S.

Kaufman

Rivara

F. P.

Bike Speed Measurements in a Recreational Population: Validity of Self Reported Speed. Injury Prevention, Vol. 3, No. 1, 1997, p. 43.

15.

Ling

A Study on Cyclist Behavior at Signalized Intersections. IEEE Transactions on Intelligent Transportation Systems, Vol. 5, No. 4, 2004, pp. 293–299.

16.

Sherony

Chien

Brink

Almutairi

Ruan

Niu

Chen

Takahashi

Development of Bicycle Carrier for Cyclist Pre-Collision System Evaluation, SAE Technical Paper 2016-01-1446, 2016. https://www.sae.org/publications/technical-papers/content/2016-01-1446/.

17.

Tian

Yang

Chien

Chen

Sherony

, eds. Estimation of the Vehicle-Pedestrian Encounter/Conflict Risk on the Road Based on TASI 110-Car Naturalistic Driving Data Collection. Proc., 2014 IEEE Intelligent Vehicles Symposium, June 8–11, Dearborn, MI, 2014.

18.

Sherony

Zhang

, eds. Pedestrian and Bicyclist Crash Scenarios in the U.S. Proc., 2015 IEEE 18th International Conference on Intelligent Transportation Systems, September 15–18, 2015; Las Palmas de Gran Canaria, Spain.

Supplementary Material

Please find the following supplemental material available below.

For Open Access articles published under a Creative Commons License, all supplemental material carries the same license as the article it is associated with.

For non-Open Access articles published, all supplemental material carries a non-exclusive license, and permission requests for re-use of supplemental material or any part of supplemental material shall be sent directly to the copyright owner as specified in the copyright notice associated with the article.

0.00 MB

0.03 MB