Comparison of Unmanned Aerial Vehicle-LiDAR and Image-Based Mobile Mapping System for Assessing Road Geometry Parameters via Digital Terrain Models

Abstract

Road condition analysis is an important research topic in many fields (such as intelligent transportation, road safety, road design analysis, and traffic analysis) and depends on road geometry parameters such as longitudinal profile and cross-slope. In this study, the extraction of road geometry parameters by unmanned aerial vehicle (UAV) with LiDAR and by a mobile photogrammetric system (MPS) designed by our research group was investigated. The purpose of this study was to obtain geometric parameters (such as road longitudinal profile and cross-slope) by using digital terrain model (DTM) surfaces derived from point cloud data acquired using UAV-LiDAR and MPS. For this purpose, a framework was developed for the extraction and comparison of longitudinal and cross-sectional profiles. First, the ground filtering approach was used to extract ground points and DTM surfaces generated from an appropriate interpolation algorithm by using ground points. Cross-sectional/longitudinal profiles of the road sections were extracted and compared with reference data. A comparison of the longitudinal profiles obtained from DTMs derived from the MPS and from UAV-LiDAR revealed root mean square error values of 1.8 cm and 2.3 cm, respectively. The average deviation of cross-slopes for both surfaces was 0.19% and 0.18%, respectively. These results show that road geometric parameters can be obtained from DTM surfaces with high accuracy. It can be concluded from the results of this study that MPS can be a favorable alternative for studies on road geometry parameters extraction.

Keywords

mobile photogrammetric system UAV-LiDAR digital terrain model slope analysis longitudinal/cross-sectional profiles road geometry parameters

Nowadays, road condition analysis has become an important topic in many fields of research, such as intelligent transportation ( 1 ), road safety, road design analysis, traffic analysis, and dynamic analysis of cars. It needs to be conducted regularly to ensure the continuity of optimum driving quality, ride comfort, and road safety to minimize road accidents. Road condition analysis depends on road geometry parameters (such as cross-slope and longitudinal profiles). These parameters provide drainage, and thus prevent the occurrence of ponding on road surfaces ( 2 ). Moreover, driving speed, tire wear, and vehicle vibration are closely related to road geometry ( 3 ). Therefore, it is important to regularly compare these parameters with the design values.

To determine road geometry parameters, various survey techniques based on accuracy specifications can be used. Surveys can be implemented with traditional techniques, such as total station (TS) ( 4 ), digital levels ( 5 ), and real-time kinematics-global navigation satellite system (RTK-GNSS) ( 6 ). Obtaining the slopes directly with TS is possible by measuring two points on the left and right sides of a road. However, when we consider the size of the area to be measured, the traffic density and the risk of accident during the surveying operation, measurements with TS are not considered practical. In addition, these methods are time consuming and labor intensive. Systems that take less time and are less labor intensive than traditional methods, for example, terrestrial laser scanning (TLS) ( 7 ), mobile laser systems (MLS) ( 2 , 8 ), airborne LiDAR systems (ALS) ( 9 ), and photogrammetry ( 10 – 12 ), have begun to be used to obtain road geometry. In addition to these systems, pavement profile scanner technology has been used to measure transverse smoothness in roadworks ( 13 ).

In the literature, LiDAR systems have generally been used in the extraction of the geometric elements of roads ( 14 – 18 ), road surface analysis ( 19 , 20 ), and detection of traffic signs on the roadside ( 21 – 23 ). To carry out these analyses, a point cloud can be used, along with a digital terrain model (DTM). Extracting information about a road from the DTM is easier and less time consuming. In this sense, Barbarella et al. ( 7 ) used TLS to conduct surveys at an airport and extracted longitudinal and transversal profiles from DTM. They reported that TS and TLS give the same level of accuracy, but TLS provides a greater amount of data. Yadav et al. ( 8 ) utilized a mobile LiDAR system to analyze road geometry and conditions. They revealed that the road width deviated by 1.4% from the reference, and the root mean square error (RMSE) was 2.1%, 2.2%, and 2.5% for the longitudinal slope, cross-slope (right), and cross-slope (left), respectively. Shams et al. ( 2 ) conducted surveys with MLS on 20 stations, including 203 travel lanes, to extract the cross-slope from DTM and tested the results statistically to evaluate the system for accuracy. They stated that mobile LiDAR is reliable for cross-slope verification with smaller than minimum acceptable accuracy value (±0.2%) at a 95% confidence level. Shams et al. ( 9 ) compared ALS and MLS systems with conventional surveying for extraction of the cross-slope. The results showed that ALS and MLS have acceptable accuracy for cross-slope verification.

Jung et al. ( 24 ) compared the TLS technique with unmanned aerial vehicle (UAV) photogrammetry and UAV-LiDAR, which have recently become popular for the roughness analysis of roads. They reported that UAV photogrammetry and UAV-LiDAR have a slight difference in cross-slope from the design value, but further research is needed to increase their accuracy for the roughness analysis of roads. Cassule et al. ( 25 ) evaluated UAV-LiDAR scanning, UAV photogrammetry, and MLS systems for how effectively these methods could extract the cross-slope. They found the observed means of measurement errors to be 0.16%, 0.29%, and 0.14% for mobile LiDAR, UAV-LiDAR, and UAV photogrammetry, respectively. Khanal et al. ( 26 ) compared the height values of the data obtained by aerial LiDAR and UAV photogrammetry methods on three different road surfaces with the results obtained by conventional methods. They achieved an accuracy of 9 cm from the aerial LiDAR system and 14 cm from the UAV photogrammetric systems. Rogers et al. ( 27 ) compared the DSM results produced by the UAV-LiDAR and UAV photogrammetry methods (different UAV models were used) in areas with different land cover. They observed that there was a difference of 0.10 m (excluding the above-vegetation) between the UAV-LiDAR system and the UAV photogrammetry. Farhadmanesh et al. ( 28 ) examined the usability of spatial data generated from mobile photogrammetry and mobile LiDAR for road bridge and pavement condition assessment. They suggested that photogrammetry could work well for a highway asset inventory system as a reliable alternative to LiDAR technology only in suitable lighting conditions, however, the LiDAR system is a better technology for vehicles moving at high speed.

With the development of technology, the use of image-based systems for mapping and extracting the road surface and its surroundings is increasing. The development of these systems and the algorithms used for processing the data obtained from them is a popular research area ( 29 , 30 ). For instance, Inzerillo et al. ( 10 ) validated low-cost and innovative techniques for the analysis of road pavements and for an assessment of their capability to improve automation and distress detection reliability with UAV-based images. They analyzed a structure-from-motion (SfM) technique, and the results showed that pavement distress can be replicated accurately. Marinelli et al. ( 11 ) investigated the extraction of information about roads (such as roadway) and the radii of the curves from images obtained with a mobile mapping vehicle on a 3.6 km route. Hu ( 12 ) conducted a study on the automatic detection of road surface defects (with 98.95% accuracy) from images obtained using a mobile photogrammetric system (MPS).

Progress in imaging sensors and photogrammetric algorithms and advances in cloud computing technology have increased interest in the development and use of low-cost image-based mobile mapping techniques and made it an attractive research topic ( 31 ). Additionally, Nebiker et al. ( 31 , 32 ) stated that such systems provide significant advantages in cost, speed, and density of information. However, most mobile mapping systems use LiDAR as the primary collection sensor for acquiring three-dimensional (3D) data, and cameras are generally used as complementary sensors to provide texture information ( 33 ). For instance, studies in the literature generally (mostly) focus on LiDAR data for the extraction of road geometry parameters ( 34 ). Nonetheless, the cost of LiDAR technology is very high and it produces intensive raw data which requires serious technical knowledge and expertise to process and analyze. LiDAR-based mobile mapping systems show limited growth in the market because of this high capacity both in cost and labor required by the technology ( 35 ). Image-based 3D reconstruction is therefore emerging as an important alternative to LiDAR systems in many application areas as a result of technological developments ( 36 ). Therefore, in studies related to road geometry, it is anticipated that image-based mobile mapping systems can be an important alternative to LiDAR-based systems.

The aim of this study is to investigate the usability and accuracy of image-based mobile mapping systems for determining road geometric parameters, where there is a gap in the available literature. In addition, the results obtained are compared with the UAV-LiDAR system and the advantages and disadvantages of both systems are examined. The aim of this study is to contribute to the literature with comprehensive analyses on the strengths and limitations of UAV-LiDAR and MPS systems for practitioners and researchers in fields such as intelligent transportation, road safety, road design analysis, and traffic analysis.

The main contributions of this paper are as follows:

The road geometric parameters can be reliably detected by the developed MPS and results also show that MPS can be a significant alternative technology for determining road geometry parameters.

We present a convenient framework to correctly and efficiently extract road geometric parameters (such as road longitudinal profile and cross-slope) via DTM surfaces created using point cloud data.

To facilitate the selection of UAV-LiDAR and MPS systems for road geometry applications, both systems are analyzed and compared in detail according to measurement time, data characteristics, data collection process, and effects of environmental conditions.

This paper is organized as follows. The study sites and data acquisition systems are explained in the next section. The proposed framework for analyzing road geometry parameters is presented in the third section. The experimental studies and results are explained in the fourth section. The final section presents discussion and conclusions.

Study Site, Data Acquisition Systems, and Dataset Description

Study Site

The experiment was conducted at Yildiz Technical University Campus, Istanbul. The study was carried out on a corridor measuring 170 m × 70 m (length: ∼170 m, slope: ∼4%, width: 15 m, as defined by the coordinates and shown in Figure 1). The area where the test study was conducted was a two-lane road section with separate inbound–outbound routes with a clear transverse and longitudinal slope change, including an alignment and a curve.

Figure 1.

Study area.

Data Acquisition Systems and Dataset Description

In this study, the data were obtained from a UAV-LiDAR system and the MPS that we designed. Figure 2 shows the MPS. The design purposes of the MPS were cost-effectiveness, data collection productivity, and ease of use. Two GoPro Hero 7 action cameras and two Topcon HyperPro dual-frequency GNSS receivers were used to form the multisensory acquisition system mounted on a van (Figure 2 shows the sensor setup). The sample frequencies at which the cameras and the GNSS receivers collected measurements were 30 FPS and 10 Hz, respectively. Table 1 shows the detailed sensor specifications. Beside the original camera videos that were acquired at 2K (2048 x 1080 pixels) resolution, frames were extracted, at 270 frames from each camera.

Figure 2.

Sensor setup: custom stereo rig with two GoPro Hero 7 cameras and two Topcon HyperPro dual-frequency global navigation satellite system (GNSS) receivers.

Table 1.

Sensor Specifications for MPS and UAV-LiDAR

	Sensor	Type	Rate	Characteristics
MPS (mobile photogrammetric system)	Camera	GoPro Hero 7	10 Hz	HyperSmooth 2704 × 1520 Linear FOV 78 Mb/s video bit rate
	Global navigationsatellite system (GNSS)	Topcon HyperPro	10 Hz	Dual-frequency (L1 + L2) multi-constellation (GPS + GLONASS)
Unmanned aerial vehicle (UAV)-LiDAR system	LiDAR	Velodyne HDL-32	700.000 Hz	35–55 mm root mean square error, 65 m range, 700,000 points/second
	GNSS	NovAtel OEM6	10 Hz	Dual-frequency (L1 + L2) multi-constellation (GPS + GLONASS + Beidou)
	Inertial measurement unit (IMU)	OEM-ADIS16488	200 Hz	Roll and pitch precision 0.04°, heading precision 0.22°

Time synchronization for multiple sensors directly affects position accuracy and algorithm performance. Therefore, a checkerboard, the coordinates of the corner points of which were known, was placed at the starting position (see Figure 3) for synchronization of camera time and GPS time. These points were accepted as ground control points (GCP) to match the GPS and camera times of collecting data at equal intervals. In brief, the first 40 consecutive frames were processed at the beginning with the SfM algorithm according to the GCP points (GCP-1, GCP-2, GCP-3, and GCP-4) that were coordinated by TS on the wall to obtain the real coordinate values of the camera poses. As a result, the actual coordinate values of the initial 40 consecutive frames were acquired, and these coordinate values were matched with the GPS coordinate values according to the following formulation, in which t represents time and P 3D position represents the correct time ( 37 ). In Equations 1, 2, 3, and 4, $t_{SFM}$ represents the video time (e.g., $t_{SFM, i}$ is the time of the i-th frame processed in the SfM procedure). To synchronize the camera information with GNSS, the time shift $τ$ is estimated in such a way that $(t_{SFM})$ + $τ$ is equal to the corresponding GPS time. P $(t_{SFM})$ is the 3D position of the GNSS receiver in the SfM reconstruction at the video time $(t_{SFM})$ . $P_{SFM (t)}$ is the GNSS position measured by the receiver at GPS time t.

P_{SFM} (t_{SFM}) \approx P_{GNSS} (t_{SFM} + τ)

(1)

P_{SFM} (t_{SFM, i}) - P_{GNSS} (t_{SFM, i} + τ) \approx 0

(2)

\hat{τ} = \arg min Σ_{i = 1}^{n} ‖ P_{SFM} (t_{SFM, i}) - P_{GNSS} (t_{SFM, i} + τ) ‖^{2}

(3)

t_{GNSS} \approx τ + t_{SFM}

(4)

Figure 3.

Checkerboard and ground control points.

The closest GNSS location to the image coordinate value obtained with the SFM algorithm was calculated, and its time value was matched using camera time (Equation 3). In addition, camera calibration was performed to obtain more accurate results. The average reprojection error was minimized in the 40 calibration images taken of a checkerboard pattern attached on a wall, with GCPs at its corner points to calibrate the intrinsic camera parameters and their relative pose. A pixel mean reprojection error of 0.38 and 0.43 was obtained for the right and left cameras, respectively.

Another experiment was carried out with the UAV-LiDAR. Figure 4 shows the UAV-LiDAR system (Phoenix Scout-32 LiDAR systems) used in this study. The UAV-LiDAR system consisted of the following components: a Velodyne laser scanner, an OEM-ADIS16488 high-precision inertial measurement unit (IMU), NovAtel OEM6 dual-frequency GNSS antennas, and a microcomputer. Table 1 shows detailed information on the sensors in the UAV-LiDAR system.

Figure 4.

Different view of Unmanned aerial vehicle (UAV)-LiDAR system used in this study (a, b, c), Reference GNSS station (d).

PPK (post-processing kinematic) was used to process the GNSS data to obtain accurate positioning results in both the MPS and the UAV-LiDAR system. Inertial Explorer post-processing software developed by NovAtel Inertial Explorer (NovAtel) was used for all processes. A reference GNSS receiver was installed within the study area to evaluate the PPK data. The Topcon Hiper Pro GNSS receiver used as the reference station is depicted in Figure 4d. The accuracy of the position data obtained from the GNSS receivers was at the level of 3–4 cm. The position information obtained by the PPK method was matched with the time tags of the images obtained from the camera to obtain the precise position information of the images ( 38 ).

In this study, the SfM method was used for generating 3D high-density points from direct georeferenced images. Post-processing of the collected MPS imagery was performed using Agisoft version 1.7.1. It involved automated point cloud densification, 3D mesh generation, digital surface modeling, and orthomosaic and digital terrain modeling ( 39 ). LiDAR data processing was done using LiDARMill software from Phoenix LiDAR Systems. First, the software combined IMU and GNSS data to create smooth and precise flight trajectories. Then, rotations and calibration models were automatically detected and eliminated. In the final stage, it took the data as a geographical reference, minimized the offsets from multiple flight lines, and exported the aligned data in the LAS format ( 40 ).

To evaluate the results, approximately 1,500 control points scattered around the road surface were measured at certain intervals on the road surface by using the geodesic GNSS receiver in the single baseline RTK mode. The measurement of these control points was carried out independently of the test studies and in accordance with the relevant studies in the literature ( 41 , 42 ). The GNSS receiver shown in Figure 4d was used as the fixed station. The coordinates of points were obtained in the ITRF05 Datum and Epoch 2005 as easting, northing, and ellipsoidal height with an accuracy of a few centimeters ( 42 – 44 ).

Methodology for Analyzing Road Geometry Parameters

The proposed framework for assessing road geometry parameters is illustrated in Figure 5. The main steps were as follows: (i) point cloud generation, (ii) ground filtering, (iii) accuracy assessment of point cloud and DTM surfaces, and (iv) longitudinal and cross-sectional profile extraction and cross-slope evaluation.

Figure 5.

Workflow of the proposed framework for assessment of road geometry.

Ground Filtering

The original point cloud should be classified as ground and non-ground to generate an accurate DTM for assessing road geometry parameters. Although there are numerous filtering algorithms, the cloth simulation filter (CSF) algorithm has attracted increasing attention in the scientific community ( 45 ). The cloth simulation algorithm was proposed by Zhang et al. ( 46 ). This algorithm starts by inverting the original point cloud data, then a virtual cloth covers the inverted point cloud. Later, the virtual cloth takes its final form as a DTM under the effects exerted by gravity and the adjacent nodes. Finally, the algorithm uses the DTM surface to distinguish the point cloud data into ground points and non-ground points ( 46 ).

The CSF algorithm is simple and easy to use. It has been successfully applied to data from different LiDAR systems (such as air, mobile, and bathymetric LiDAR) and implemented using numerous software programs ( 45 – 47 ). In this study, the ground points were detected and extracted from raw point cloud data obtained by UAV-LiDAR and MPS by using the CSF algorithm to generate DTM surfaces, and the road geometry parameters were evaluated using these DTMs.

Modeling of the Filtered Point Cloud

The evaluation of road geometry was realized using a DTM generated by the interpolation of point cloud data. Because of the significant advantages in simplicity, computation time, and data storage, longitudinal and cross-sectional profiles were extracted directly from the DTM surfaces. In addition, the height and slope values of these profiles were calculated from a numerical model of road surfaces.

Point cloud data obtained from both UAV-LiDAR and MPS consist of large datasets that contain millions of points; therefore, the chosen interpolation method for the generation of a DTM surface should be fast for intensive data and represent a surface well by detecting small changes. Test studies for different algorithms in the literature have revealed that the most appropriate method is inverse distance to power (IDW). In our study, this method was used as the interpolation method ( 2 , 7 ).

To determine the appropriate grid resolution size, we utilized the formula suggested by Hengl ( 48 ). This formula allows for the calculation of the minimum resolution value (p) depending on the relationship between the data density and the grid resolution, as given in Equation 5.

p = 0.5 \sqrt{\frac{1}{D}}

(5)

where D is the average density of the point cloud.

Accuracy Assessment of Point Cloud and DTM Surfaces

The accuracy assessment of point clouds and DTM surfaces consists of two stages and involves the evaluation of the relative accuracy between point clouds and the absolute accuracy between DTM surfaces and the manually measured RTK-GNSS control points. In this study, to determine the relative accuracy of point clouds, a multiscale model-to-model cloud comparison (M3C2) algorithm was used ( 49 ). In the transportation corridor, the topography of road sections is generally flat and has a slight variation in slope, and this inconsistency mostly occurs in the vertical direction. To be compatible with the purpose of our study, therefore, vertical differences were calculated and examined ( 41 ). Thus, the M3C2 algorithm was utilized, as it can work directly on point cloud data and produces reliable results to determine the accuracy of the point cloud. The M3C2 algorithm starts with the calculation of the surface normal in 3D on the basis of the detected core points. The surface distance is calculated between two point clouds along the local surface direction ( 50 ). In this study, the M3C2 plug-in provided by CloudCompare was used to detect the compatibility between UAV-LiDAR and MPS point cloud data ( 51 ).

The absolute accuracy of the point cloud data was achieved by comparing the manually measured RTK-GNSS points and the generated DTM surface from the filtered point clouds. Therefore, the mean error (ME), RMSE, mean absolute error (MAE), and the standard deviation of error (SDE) were computed using the differences in elevation between the measured RTK-GNSS check points (Z_RTK-GNSS) and those at the same location of the DTM surface (Z_DTM) ( 42 , 52 , 53 ). The ME, MAE, RMSE, and SDE were calculated as follows:

Δ Z_{i} = Z_{RTK - GNSS} (x_{i}, y_{i}) - Z_{DTM} (x_{i}, y_{i})

(6)

Root - Mean - Square - Error = \sqrt{\frac{\sum_{i = 1}^{n} {(Δ Zi)}^{2}}{n}}

(7)

Mean Error = \frac{\sum_{i}^{n} (Δ Zi)}{n}

(8)

Mean Absolute Error = \frac{\sum_{i}^{n} | (Δ Zi) |}{n}

(9)

Standard Deviation of Error = \sqrt{\frac{\sum_{i}^{n} {(Δ Zi - ME)}^{2}}{n - 1}}

(10)

Extraction and Comparison of Longitudinaland Cross-Sectional Profiles

In this study, we aimed to investigate whether UAV-LiDAR or MPS could be used to assess road geometry. The longitudinal and cross-sectional profiles of road pavement surfaces were extracted from datasets obtained from both systems to analyze the elevation and slope changes within these profiles according to the reference data, then the accuracy was determined. The longitudinal profile could be expressed as an intersection line parallel to the road surface along the direction of the lane, and a cross-sectional profile could be defined as the line perpendicular to the road surface. As indicated above, the profiles were derived from the DTM surfaces created from the point cloud data obtained from both UAV-LiDAR and MPS. Accurate reference data were needed to perform the intended analyses effectively. Therefore, planimetric coordinates and elevation values of the points that formed the reference longitudinal profiles were measured with the RTK-GNSS technique at an average of 4 m intervals along the A–B and C–D road section lines corresponding to the centerline of the up and down direction of the road, as shown in Figure 6. In contrast, the reference points that formed the cross-sections were measured with the RTK-GNSS technique perpendicular to the A–B and C–D road axes, as shown in Figure 6.

Figure 6.

Shaded relief map of grid digital terrain model (DTM). Red dots show longitudinal profile points in the A–B road segment, blue dots show longitudinal profile for C–D road segment, cyan dots show cross-sections: (a, b) visualization of longitudinal profile, (c, d) some characteristics of cross-sectional profiles.

For UAV-LiDAR and MPS-based data, the elevations of longitudinal points along the A–B and C–D sections were extracted from the grid DTM by using bilinear interpolation ( 7 ). Then, the elevation differences between the reference (Z_Ref-Long) and the DTM-derived (Z_DTM-Long) profiles were calculated, and the ME, MAE, RMSE, and SDE values, the formulas of which were given in the previous section, were estimated to determine the accuracy of the two systems.

Δ Z_{i} = Z_{Ref - Long} (x_{i}, y_{i}) - Z_{DTM - Long} (x_{i}, y_{i})

(11)

The next step was to measure and calculate the cross-slope values of 10 different transversal sections identified along the road line, as shown in Figure 6. For the performance accuracy analysis, the reference cross-slope values of these determined sections and the cross-slope values obtained from the DTM surfaces were compared. For this purpose, reference elevation values were created by measuring separately between pavements or lane lines along the A–B and C–D road routes using the RTK-GNSS technique. Elevation data for the same segments were also obtained from the DTM surfaces. Thereafter, linear regression was applied between the extracted/measured elevation value and the transverse width of the road section. Thus, the cross-slope value was obtained from the slope of the regression line ( 14 ). These procedures were developed and implemented in the MATLAB environment.

Experimental Results

In this section, the experimental results are explained step by step. First, the results of the ground-point classification and both the relative and the absolute accuracy of these data are explained in detail. Then, the analyses carried out to obtain and compare road geometry parameters from the data obtained from both systems are discussed.

Results of Ground-Point Filtering

The UAV-LiDAR and MPS point clouds for our test areas were filtered using the CSF algorithm to detect ground points. For the CSF algorithm, it is essential to determine the cloth resolution (C.R), maximum iteration (M.I), and classification threshold (C.T) parameters. The parameters of the CSF algorithm were determined through many trials as, respectively, 0.3, 500, and 0.4 for the UAV-LiDAR data and 0.3, 200, and 0.7 for the MPS data; a flat surface was chosen as the scene type for both datasets. As a result of the filtering study, special attention was paid to the detection and cleaning of the objects (vehicle, pedestrian, sign, tree, and lamppost) placed on and around the road surface to perform the road geometric analysis more precisely and accurately. The filtering results were manually verified and corrected by detecting the misclassified points, thus the final ground-point data were determined.

Figure 7, a and b , shows the raw point cloud data obtained with both systems, and Figure 7, c and d , indicates the ground points obtained as a result of filtering the same regions with the CSF algorithm. In both sets of point cloud data, different non-ground objects (such as trees, lampposts, vehicles, and pedestrians) were successfully filtered, and the ground points were detected effectively.

Figure 7.

Comparison between raw and filtered point clouds before and after the cloth simulation filter (CSF) algorithm. Raw point cloud data: (a) Unmanned aerial vehicle (UAV)-LiDAR, (b) mobile photogrammetric system (MPS). Filtered point cloud data with CSF algorithm: (c) UAV-LiDAR, (d) MPS.

Accuracy Assessment of Point Cloud and DTM Surfaces

Assessment of the accuracy of point cloud and DTM surfaces involves examining both relative and absolute accuracy. For relative accuracy, the discrepancy between UAV-LiDAR and MPS point clouds was determined using the M3C2 method. Figure 8 shows the surface obtained as a result of comparing the two different datasets. In this context, Figure 8, a and b , shows the M3C2 distances obtained by comparing two different datasets, and c represents the distribution of the M3C2 distances as a histogram. The maximum and minimum M3C2 distances of the surface (Figure 8a) were 10 cm and −15 cm. Further, the mean and standard deviation of the M3C2 distance were determined to be −2.3 cm and 2.2 cm, respectively. The histogram of the calculated M3C2 distances was a concentration within ±6 cm and shows normal distribution characteristics. When Figure 8a is examined, it has been observed that the erroneous points are generally concentrated in small and limited regions such as the road borders or the median edge, and it has been interpreted that these errors can be caused by the lack or inconsistency of the data in those regions. However, when the study area was examined in general, it was seen that the green area was much denser and approximately more than 95% of the calculated M3C2 distance was in the range of ±6 cm. In general, based on these results, both systems exhibited a good degree of agreement in calculated discrepancy and were found to be comparable.

Figure 8.

Comparison of unmanned aerial vehicle (UAV)-LiDAR and mobile photogrammetric system (MPS) using the multiscale model-to-model cloud comparison (M3C2) method: (a, b) M3C2 distances of two different point clouds. (c) M3C2 distance histogram with mean and standard deviation of M3C2 distance.

The absolute accuracy of the DTM surfaces generated from two different point clouds was examined against the RTK-GNSS checkpoints. Figure 9 shows the shaded relief maps of the gridded DTM for UAV-LiDAR and MPS, with the RTK-GNSS points on the top of the DTM surfaces. Figure 9 shows that 1,500 RTK-GNSS points were measured to cover the entire DTM surface to determine the absolute vertical accuracy of the DTMs as precisely as possible. Error statistics for the DTM surfaces against the measured RTK-GNSS checkpoints are shown in Table 2. As seen in Table 2, the ME, MAE, RMSE, and SDE values were below centimeter level. The error statistics obtained for both systems demonstrate that their vertical accuracy in the DTMs was very similar. Furthermore, results revealed that the vertical accuracy of both DTM surfaces was very high, and the surfaces were satisfactorily georeferenced to centimeter scale for the purpose of this study.

Figure 9.

Shaded relief map of the grid digital terrain model (DTM): (a) unmanned aerial vehicle (UAV)-LiDAR and (b) mobile photogrammetric system (MPS). Red dots are the real-time kinematics-global navigation satellite system (RTK-GNSS) surveyed points for absolute vertical accuracy.

Table 2.

Error Statistics for Modeled Digital Terrain Model (DTM) Against Measured Real-Time Kinematics-Global Navigation Satellite System (RTK-GNSS) Checkpoints

DTM surfaces	Mean error (m)	Root mean square error (m)	Standard deviation error (m)	Mean absolute error (m)
Unmanned aerial vehicle-LiDAR	0.022	0.056	0.053	0.025
Mobile photogrammetric system	−0.001	0.052	0.051	0.021

Comparative Performance Analysis of Longitudinal and Cross-Sectional Profiles Based on Elevation and Cross-Slope Values

Figure 10, a–c, shows a comparison of the longitudinal profiles obtained along both the A–B and the C–D sections, with elevation values from the reference data and the UAV-LiDAR and MPS-derived DTM surfaces. As shown in Figure 10, a–c, the dashed black line, blue circle, and orange plus sign represent longitudinal profiles generated from reference, MPS, and UAV-LiDAR-derived DTM, respectively. Figure 10, b–d, shows the height difference between the reference longitudinal profile and profiles generated from UAV-LiDAR and MPS-derived DTM. In addition, the distribution of the residuals calculated for the elevation values of the longitudinal profiles between the reference and DTMs using Equation 10 is illustrated in Figure 11. The calculated ME, MAE, RMSE, and SDE values for these residuals are shown in Table 3. When the differences between the reference and the two DTM-generated longitudinal profiles in the A–B section were examined, although the elevations of the profiles created from the both DTMs showed small deviations in very limited areas, the profiles were highly compatible with each other. Moreover, the ME, MAE, RMSE, and SDE values obtained from both DTMs remained at the centimeter level. An average deviation of 1 cm was observed when these values were compared among themselves, and slightly better results were acquired with MPS-derived DTM for RMSE, MAE, and ME values. Similarly, the variance of the elevation differences was nearly the same for both DTMs, but the median value was close to zero for MPS-derived DTM, while it was slightly larger for the UAV-LiDAR-derived DTM.

Figure 10.

Comparison of different longitudinal profiles generated from reference, unmanned aerial vehicle (UAV)-LiDAR-derived, and mobile photogrammetric system (MPS)-derived digital terrain model (DTM) (a–c), with height difference between the reference longitudinal profile and profiles generated from UAV-LiDAR- and MPS-derived DTM (b–d).

Figure 11.

Statistics of the elevation difference between the reference points of the longitudinal profile and the digital terrain model-generated values for the same position for (a) A–B section and (b) C–D section with residual plots of the 25th percentile, 75th percentile, and the median.

Table 3.

Error Statistics Values of Longitudinal Profile Differences

Road sections	Comparison with reference profiles	Mean error (m)	Root mean square error (m)	Standard deviation error (m)	Mean absolute error (m)
A–B longitudinal profile	UAV-LiDAR	0.017	0.025	0.018	0.022
	MPS	−0.001	0.016	0.021	0.013
C–D longitudinal profile	UAV-LiDAR	0.016	0.022	0.012	0.019
	MPS	−0.012	0.02	0.032	0.017

Note: UAV = unmanned aerial vehicle; MPS = mobile photogrammetric system.

An examination of the profiles and elevation differences between the reference and the DTMs in the C–D sections revealed similar characteristics to those of the A–B road section. However, the ME, MAE, RMSE, and SDE values for the C–D section were lower for UAV-LiDAR-derived DTM, while slightly worse results were obtained for MPS-derived DTM. Nonetheless, the results of both longitudinal profiles were very close to one another and the curves showed a high level of agreement. These results showed that the DTM surfaces obtained from the UAV-LiDAR and MPS point cloud had similar high accuracy and that the developed MPS could potentially be used for the extraction of road geometry parameters.

For both the road sections (A–B and C–D), the reference cross-slope values were calculated from the RTK-GNSS data, and these reference values were compared with the cross-slope values calculated from the UAV-LiDAR- and MPS-derived DTMs. As shown in Table 4, in the A–B road section, the cross-slope difference between the reference data and the UAV-LiDAR-derived DTM ranged from 0.01% to 0.33%, and the average difference was 0.15%. For the same road section, the difference in the cross-slope between the reference data and the MPS-derived DTM ranged from 0.04% to 0.46%, with an average difference of 0.17%. The cross-slope differences for the C–D road segment ranged from 0.02% to 0.36%, and the average was 0.2% for the UAV-LiDAR-derived DTM, while it ranged from 0% to 0.41, with an average of 0.2%, for the MPS-derived DTM. According to the relevant literature and technical guidelines, it is generally defined as 0.2% of the acceptable accuracy for cross-slope differences ( 2 , 14 , 54 ). In this study, the average difference values obtained with both DTMs were below this limit, and these systems yielded promising results for the cross-slope evaluation.

Table 4.

Cross-Slope Comparison Between Ground Truth and Digital Terrain Models Derived Using UAV-LiDAR and MPS

A–B Section				C–D Section
	Ground truth (%)	Difference from ground truth		Ground truth (%)	Difference from ground truth
	Ground truth (%)	UAV_LiDAR_DTM (%)	MPS_DTM (%)	Ground truth (%)	UAV_LiDAR_DTM (%)	MPS_DTM (%)
Sec_1	0.53	0.1	0.09	2.55	0.22	0.31
Sec_2	0.47	0.33	0.46	0.71	0.25	0.41
Sec_3	1.16	0.06	0.2	0.12	0.22	0.1
Sec_4	0.27	0.15	0.05	1.15	0.21	0.27
Sec_5	1.56	0.23	0.07	0.52	0.21	0.41
Sec_6	1.12	0.28	0.24	1.28	0.26	0.16
Sec_7	1.47	0.01	0.04	0.47	0.3	0
Sec_8	1.34	0.1	0.13	0.7	0.03	0.17
Sec_9	1.19	0.15	0.26	0.88	0.02	0.1
Sec_10	1.31	0.07	0.19	1.27	0.36	0.08
Mean	NA	0.15	0.17	NA	0.2	0.2

Note: UAV = unmanned aerial vehicle; DTM = digital terrain model; MPS = mobile photogrammetric system; NA= not available.

Discussion and Conclusion

In this study, the extraction of road geometry parameters with the UAV-LiDAR system, which has recently become increasingly popular ( 55 ), and the MPS ( 37 ) designed by our research group was investigated. In this context, a cost-effective MPS system consisting of GNSS and an action camera was designed and used along with the UAV-LiDAR system, and the data were gathered in a road environment with different slopes and heights. The data obtained from both systems were compared with the reference data measured by RTK-GNSS, and their accuracy was analyzed. A comparison of the longitudinal profiles obtained from the MPS-derived and UAV-LiDAR-derived DTMs with the reference profile elevations revealed RMSE values of 1.8 cm and 2.3 cm, respectively.

Furthermore, comparisons between the cross-slope values calculated for the cross-sections from the DTMs and the reference value revealed that the average deviations were 0.19% and 0.18%, respectively. All these results showed that MPS is a convenient alternative for road geometry parameter extraction studies and it can be considered an innovative approach. In addition, high-resolution DTM surfaces created using dense point clouds provided significant advantages in the extraction of road geometric parameters and offered the opportunity to obtain geometric information on the surface at any time quickly.

Many different studies have been carried out in the literature to analyze road geometric parameters. When our results were compared with other studies in the literature, it was seen that there were some similarities and differences. In this context, the results of Shams et al. ( 2 ) and Tsai et al. ( 14 ) were better than the accuracy threshold value (±0.2%) in studies on cross-slope extraction. Similarly, Gargoum et al. ( 17 ) found cross-slope values between 0.08% and 0.22%. Yadav et al. ( 8 ) found that the RMSE values were 2.2% and 2.5% for cross-slope (right) and cross-slope (left), respectively. It can be seen that these results were similar to the results we obtained. However, according to a study conducted by Cassule et al. ( 25 ), the UAV-LiDAR system gave lower results in finding the cross-slope value compared with UAV photogrammetry and mobile LiDAR. On the contrary, in our study, the cross-slope values determined from the UAV-LiDAR system were better than MPS and were very close to the reference values. This difference between the two studies might arise from the laser system, point density, or methodology used. In the studies carried out for the extraction of longitudinal profile, Yadav et al. ( 8 ) found the RMSE value of 0.2%, and Frutos and Castro ( 56 ) obtained the geometric elements related to the longitudinal profile with an error of less than 8 cm. In our study, results with higher accuracy (RMSE values of 1.8 cm and 2.3 cm for MPS- and UAV-LiDAR-derived DTMs, respectively) were obtained for the longitudinal profiles of the road.

MPS and UAV-LiDAR systems have advantages and disadvantages when compared with each other. The advantage of the UAV-LiDAR system over MPS is that it is not affected by environmental factors (such as lighting conditions, characteristics of the photographed object, weather and light conditions). UAV-LiDAR is a compact system with a data processing time that is significantly shorter than that of MPS based on the SfM algorithm and it can also directly generate 3D coordinates. Its disadvantages are that it is affected by canopy cover (such as buildings, trees, overpass, tunnels) that will restrict the view of the road surface and it requires extensive planning before measurement. Besides, it produces relatively low point density compared with MPS and the costs of software and hardware are higher. Since the UAV-LiDAR system has flight capability of just 20–30 min, it is not useful for long-distance roadworks. The most important advantages of the MPS system compared with the UAV-LiDAR system are that it can be used over long distances and its low investment cost. Moreover, MPS outperformed the UAV-LiDAR system with its ability to obtain detailed data about the road environment, in addition to the homogeneous and dense point cloud.

For this study, a standard road route with a low traffic load that is located within the campus was selected. As a result of the studies carried out on the selected route, it was observed that the proposed system can be a competitive alternative to other systems. However, MPS performance may vary for different road types and situations. Moving and stationary objects (pedestrians, vehicles, etc.) of very different types and sizes, especially on roads in urban areas, will make it difficult to obtain smooth and sensitive image data of the road surface and its surroundings with an MPS system. Similarly, it is thought that the structure and type of the road will affect the performance of the proposed system. For example, the proposed system may not achieve the desired data quality with its current configuration when applied to the high number of lanes and speed limits on highways. In these cases, the configuration of the proposed system and the height, position, angle, number, and so forth of the camera should be modified according to the road type. In future studies (across different road types, conditions, and imaging sensors), we plan to carry out comprehensive tests using various imaging systems in different road types and conditions and to generalize the proposed system.

We anticipate that this study will contribute significantly to the literature as the MPS technology is being used for the first time to determine road geometric parameters. Moreover, it was concluded that MPS can be an important alternative to different LiDAR systems, conventional surveying techniques, or both, that are frequently used to determine road geometries. It is thought that this study will be an important base and guideline for researchers studying this subject. Also, considering the potential of image-based systems to extract road geometry, in future case studies it would be interesting to add the DTM from UAV photogrammetry, to evaluate the results from vertical (UAV) and oblique (mobile mapping) SfM point clouds.

Footnotes

Author Contributions

The authors confirm contribution to the paper as follows: study conception and design: M. Gurturk, B. Suleymanoglu; data collection: M. Gurturk, B. Suleymanoglu, Y. Yilmaz, M. Soycan ; analysis and interpretation of results: M. Gurturk, B. Suleymanoglu, Y. Yilmaz, M. Soycan; draft manuscript preparation: M. Gurturk, B. Suleymanoglu, Y. Yilmaz, M. Soycan, A. Soycan. 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: This research was supported by the project number FDK-2019-3593 and FDK-2019-3597, which was accepted by the Yildiz Technical University Scientific Research Projects Commission.

ORCID iDs

Mert Gurturk

Yalcin Yilmaz

References

Pradhan

Ibrahim

Road Geometric Modeling Using a Novel Hierarchical Approach. In Laser Scanning Systems in Highway and Safety Assessment, Springer, Berlin/Heidelberg, Germany, 2020, Cham, 2020, pp. 33–46.

Shams

Sarasua

W. A.

Famili

Davis

W. J.

Ogle

J. H.

Cassule

Mammadrahimli

Highway Cross-Slope Measurement Using Mobile LiDAR. Transportation Research Record: Journal of the Transportation Research Board, 2018. 2: 175–199.

De Blasiis

M. R.

Di Benedetto

Fiani

Garozzo

Assessing of the Road Pavement Roughness by Means of Lidar Technology. Coatings, Vol. 11, 2021, p. 17.

Mill

Ellmann

Uueküla

Joala

Road Surface Surveying Using Terrestrial Laser Scanner and Total Station Technologies. Proc., 8th International Conference on Environmental Engineering, Vol. 8, 2011, Vilnius, Lithunia, pp. 1142–1146.

Wang

Cross Slope: Revised Specification Highlights. State Construction Office, Department of Transportation, Tallahassee, FL, 2010.

Gurturk

Yılmaz

Suleymanoglu

Soycan

High Definition Corridor Mapping From Images Sequences. International Journal of Digital Innovation in the Built Environment, Vol. 9, No. 1, 2020, pp. 22–35.

Barbarella

De Blasiis

M. R.

Fiani

Terrestrial Laser Scanner for the Analysis of Airport Pavement Geometry. International Journal of Pavement Engineering, Vol. 20, No. 4, 2019, pp. 466–480.

Yadav

Singh

A. K.

Lohani

Computation of Road Geometry Parameters Using Mobile LiDAR System. Remote Sensing Applications: Society and Environment, Vol. 10, 2019, pp. 18–23.

Shams Esfandabadi

Highway Cross Slope Measurement Using Airborne and Mobile LiDAR. PhD thesis. Clemson University, South Carolina, 2018.

10.

Inzerillo

Di Mino

Roberts

Image-Based 3D Reconstruction Using Traditional and UAV Datasets for Analysis of Road Pavement Distress. Automation in Construction, Vol. 96, 2018, pp. 457–469.

11.

Marinelli

Bassani

Piras

Lingua

A. M.

Mobile Mapping Systems and Spatial Data Collection Strategies Assessment in the Identification of Horizontal Alignment of Highways. Transportation Research Part C: Emerging Technologies, Vol. 79, 2017, pp. 257–273.

12.

Degenerate Near-Planar Road Surface 3D Reconstruction and Automatic Defects Detection. PhD thesis. Virginia Tech, Virginia, 2020.

13.

von Olshausen

Roetner

Koch

Reiterer

Multimodal Measurement System for Road Analysis and Surveying of Road Surroundings. Proc., Automated Visual Inspection and Machine Vision IV, Vol. 11787, Bellingham, WA, 2021, pp. 72–78.

14.

Tsai

Wang

Pitts

Mobile Cross-Slope Measurement Method Using Lidar Technology. Transportation Research Record: Journal of the Transportation Research Board, 2013. 2367: pp. 53–59.

15.

Hui

Jin

Yevenyo

Y. Z.

Road Centerline Extraction From Airborne LiDAR Point Cloud Based on Hierarchical Fusion and Optimization. ISPRS Journal of Photogrammetry and Remote Sensing, Vol. 118, 2016, pp. 22–36.

16.

Zeybek

Biçici

Geometric Feature Extraction of Road From UAV Based Point Cloud Data. Proc., Third International Conference on Smart City Applications, Springer, Cham, Switzerland, 2020, pp. 435–449.

17.

Gargoum

S. A.

El-Basyouny

Froese

Gadowski

A Fully Automated Approach to Extract and Assess Road Cross Sections From Mobile LiDAR Data. IEEE Transactions on Intelligent Transportation Systems, Vol. 19, No. 11, 2018, pp. 3507–3516.

18.

Wang

Chapman

M. A.

Mobile Laser Scanned Point-Clouds for Road Object Detection and Extraction: A Review. Remote Sensing, Vol. 10, No. 10, 2018, p. 1531.

19.

Yadav

Singh

A. K.

Lohani

Extraction of Road Surface From Mobile LiDAR Data of Complex Road Environment. International Journal of Remote Sensing, Vol. 38, No. 16, 2017, pp. 4645–4672.

20.

Husain

Vaishya

R. C.

Road Surface and its Center Line and Boundary Lines Detection Using Terrestrial Lidar Data. The Egyptian Journal of Remote Sensing & Space Sciences, Vol. 21, No. 3, 2018, pp. 363–374.

21.

Weng

Chen

Wang

Road Traffic Sign Detection and Classification From Mobile LiDAR Point Clouds. Proc., 2nd ISPRS International Conference on Computer Vision in Remote Sensing, Xiamen, China, 2016, pp. 55–61.

22.

Gargoum

El-Basyouny

Automated Extraction of Road Features Using LiDAR Data: A Review of LiDAR Applications in Transportation. Proc., 2017 4th International Conference on Transportation Information and Safety, ICTIS 2017, Banff, AB, Canada, 2017, pp. 563–574.

23.

Gargoum

El-Basyouny

Effects of LiDAR Point Density on Extraction of Traffic Signs: A Sensitivity Study. Transportation Research Record: Journal of the Transportation Research Board, 2019. 2673: 41–51.

24.

Jung

K. Y.

Park

J. K.

Roughness Analysis of Paved Road Using UAV-LiDAR and Images. Journal of the Korean Society of Surveying, Geodesy, Photogrammetry and Cartography, Vol. 39, No. 1, 2021, pp. 55–63.

25.

Cassule

Fooladi-Mahani

Burgett

Shams

Famili

Combined Use of LiDAR and Photogrammetry for Highway Cross-Slope Measurement, Clemson, 2021.

26.

Khanal

Hasan

Sterbentz

Johnson

Weatherly

Accuracy Comparison of Aerial Lidar, Mobile-Terrestrial Lidar, and UAV Photogrammetric Capture Data Elevations Over Different Terrain Types. Infrastructures, Vol. 5, No. 8, 2020, p. 65.

27.

Rogers

S. R.

Manning

Livingstone

Comparing the Spatial Accuracy of Digital Surface Models From Four Unoccupied Aerial Systems: Photogrammetry Versus LiDAR. Remote Sensing, Vol. 12, No. 17, 2020, p. 2806.

28.

Farhadmanesh

Cross

Mashhadi

A. H.

Rashidi

Wempen

Highway Asset and Pavement Condition Management Using Mobile Photogrammetry. Transportation Research Record: Journal of the Transportation Research Board, 2021. 2675: 296–307.

29.

Simultaneous Three-Dimensional Mapping and Geolocation of Road Surface. PhD thesis. Virginia Tech, Virginia, 2018.

30.

Tapken

Mobile Mapping Trends and Insights. GIM International. The Worldwide Magazine for Geomatics, Vol. 32, No. 4, 2018, pp. 21–24.

31.

Nebikera

Muttenz

S. S.

3D Imagery for Infrastructure Management–Mobile Mapping Meets the Cloud. Proc., 56th Photogrammetric Week, Stuttgart, Germany, 2017.

32.

Nebiker

Cavegn

Loesch

Cloud-Based Geospatial 3D Image Spaces—A Powerful Urban Model for the Smart City. ISPRS International Journal of Geo-Information, Vol. 4, No. 4, 2015, pp. 2267–2291.

33.

Elhashash

Albanwan

Qin

A Review of Mobile Mapping Systems: From Sensors to Applications. Sensors, Vol. 22, No. 11, 2022, p. 4262.

34.

Gargoum

S. A.

Basyouny

K. El.

A Literature Synthesis of LiDAR Applications in Transportation: Feature Extraction and Geometric Assessments of Highways. GIScience & Remote Sensing, Vol. 56, No. 6, 2019, pp. 864–893.

35.

Balado

González

Arias

Castro

Novel Approach to Automatic Traffic Sign Inventory Based on Mobile Mapping System Data and Deep Learning. Remote Sensing, Vol. 12, No. 3, 2020, p. 442.

36.

Biçici

Zeybek

Effectiveness of Training Sample and Features for Random Forest on Road Extraction From Unmanned Aerial Vehicle-Based Point Cloud. Transportation Research Record: Journal of the Transportation Research Board, 2021. 2675: 401–418.

37.

Gurturk

Yusefi

Aslan

M. F.

Soycan

Durdu

Masiero

The YTU Dataset and Recurrent Neural Network Based Visual-Inertial Odometry. Measurement, Vol. 184, No. 13, 2021, p. 109878.

38.

Gurturk

Soycan

Accuracy Assessment of Kinematic PPP Versus PPK for GNSS Flights Data Processing. Survey Review, Vol. 54, No. 382, 2022, pp. 48–56.

39.

Sagan

Maimaitijiang

Sidike

Eblimit

Peterson

K. T.

Hartling

Esposito

, et al. UAV-Based High Resolution Thermal Imaging for Vegetation Monitoring, and Plant Phenotyping Using ICI 8640 P, FLIR Vue Pro R 640, and Thermomap Cameras. Remote Sensing Vol. 11, No. 3, 2019, p. 330.

40.

Maimaitijiang

Sagan

Erkbol

Adrian

Newcomb

Lebauer

Pauli

Shakoor

Mockler

T. C.

UAV-Based Sorghum Growth Monitoring: A Comparative Analysis of Lidar and Photogrammetry. Proc., ISPRS Annals of the Photogrammetry, Remote Sensing and Spatial Information Sciences, Vol. 5, 2020, pp. 489–496.

41.

Lin

Y. C.

Manish

Bullock

Habib

Comparative Analysis of Different Mobile LiDAR Mapping Systems for Ditch Line Characterization. Remote Sensing Vol. 13, No. 13, 2021, p. 2485.

42.

Tamminga

A. D.

Eaton

B. C.

Hugenholtz

C. H.

UAS-Based Remote Sensing of Fluvial Change Following an Extreme Flood Event. Earth Surface Processes and Landforms, Vol. 40, No. 11, 2015, pp. 1464–1476.

43.

Dabove

The Usability of GNSS Mass-Market Receivers for Cadastral Surveys Considering RTK and NRTK Techniques. Geodesy and Geodynamics, Vol. 10, No. 4, 2019, pp. 282–289.

44.

Bakula

Constrained and Network Multi-Receiver Single-Epoch RTK Positioning. Survey Review, Vol. 52, No. 373, 2020, pp. 289–298.

45.

Cai

Zhang

Liang

Wan

Yan

Shao

Filtering Airborne LiDAR Data Through Complementary Cloth Simulation and Progressive TIN Densification Filters. Remote Sensing, Vol. 11, No. 9, 2019, p. 1037.

46.

Zhang

Wan

Wang

Xie

Wang

Yan

An Easy-to-Use Airborne LiDAR Data Filtering Method Based on Cloth Simulation. Remote Sensing, Vol. 8, No. 6, 2016, p. 501.

47.

Yang

Zhao

Filtering of Airborne LiDAR Bathymetry Based on Bidirectional Cloth Simulation. ISPRS Journal of Photogrammetry and Remote Sensing, Vol. 163, 2020, pp. 49–61.

48.

Hengl

Finding the Right Pixel Size. Computers & Geosciences, Vol. 32, No. 9, 2006, pp. 1283–1298.

49.

Lague

Brodu

Leroux

Accurate 3D Comparison of Complex Topography With Terrestrial Laser Scanner: Application to the Rangitikei Canyon (NZ). ISPRS Journal of Photogrammetry and Remote Sensing, Vol. 82, 2013, pp. 10–26.

50.

DiFrancesco

P. M.

Bonneau

Hutchinson

D. J.

The Implications of M3C2 Projection Diameter on 3D Semi-Automated Rockfall Extraction From Sequential Terrestrial Laser Scanning Point Clouds. Remote Sensing, Vol. 12, No. 11, 2020, p. 1885.

51.

Girardeau-Montaut

CloudCompare (Version 2.11 Alpha). 2020. http://www.cloudcompare.org. Accessed June 4, 2020.

52.

Smith

M. W.

Carrivick

J. L.

Hooke

Kirkby

M. J.

Reconstructing Flash Flood Magnitudes Using “Structure-From-Motion”: A Rapid Assessment Tool. Journal of Hydrology, Vol. 519, 2014, pp. 1914–1927.

53.

Williams

R. D.

Brasington

Hicks

Measures

Rennie

C. D.

Vericat

Hydraulic Validation of Two-Dimensional Simulations of Braided River Flow With Spatially Continuous aDcp Data. Water Resources Research, Vol. 49, No. 9, 2013, pp. 5183–5205.

54.

Hunt

J. E.

Vandervalk

Snyder

SHRP2 Report S2-S03-RW-01: Roadway Measurement System Evaluation. Transportation Research Board of the National Academies, Washington, D.C., 2013.

55.

Outay

Mengash

H. A.

Adnan

Applications of Unmanned Aerial Vehicle (UAV) in Road Safety, Traffic and Highway Infrastructure Management: Recent Advances and Challenges. Transportation Research: Part A Policy and Practice, Vol. 141, 2020, pp. 116–129.

56.

Higuera de Frutos

Castro

A Method to Identify and Classify the Vertical Alignment of Existing Roads. Computer-Aided Civil and Infrastructure Engineering, Vol. 32, No. 11, 2017, pp. 952–963.