Ahmed Tageldin

Application of Computer Vision to Diagnosis of Pedestrian Safety Issues

The potential for using computer vision techniques to solve several shortcomings associated with traditional road safety and behavior analysis is demonstrated. Surrogate data such as traffic conflicts provide invaluable information that can be used to understand collision-contributing factors and the collision failure mechanism better. Recent advances in computer vision techniques have encouraged the use of proactive safety surrogate measures such as detection of conflicts and violations. The objective of this study is to demonstrate the automated safety diagnosis of pedestrian crossing safety issues by using computer vision techniques. The automated safety diagnosis is applied at a major signalized intersection in downtown Vancouver, British Columbia, Canada, at which concerns had been raised regarding the high conflict rate between vehicles and pedestrians as well as the elevated number of traffic violations (i.e., jaywalking). This study is unique in its attempt to extract conflict indicators and detect violations from video sequences in a fully automated way. This line of research benefits safety experts because it provides a prompt and objective safety evaluation for intersections. The research also provides a permanent database for traffic information that can be beneficial for a sound safety diagnosis as well as for developing safety countermeasures.

Can Time Proximity Measures Be Used as Safety Indicators in All Driving Cultures? Case Study of Motorcycle Safety in China

Limitations associated with traditional collision-based safety analysis techniques led to a growing interest in the use of surrogate safety measures such as the traffic conflict technique. This interest was facilitated by advances in automated video-based data collection methods that helped to overcome the reliability issues associated with manual collection of data on traffic conflicts. Various objective conflict indicators that measure various spatial and temporal aspects of user proximity are available to measure the severity of traffic events. These time-proximity conflict measures assume that proximity is a surrogate for conflict severity. However, this assumption may not be valid in many driving environments. The objective of this paper is to investigate whether time-proximity conflict measures can be a good indicator of safety in less-organized traffic environments with highly mixed road users. A case study of motorcycle conflicts in a highly congested shared intersection in Shanghai, China, was used as a case study. Traffic conflicts were analyzed with the use of automated video-based analysis techniques. Several traffic conflict indicators designed to detect evasive actions, such as deceleration, jerk, and yaw rate, were recommended as better able to measure traffic conflicts in such traffic environments. The results showed that indicators that measured evasive actions had higher potential to identify motorcycle conflicts in highly mixed, less-organized traffic environments than did time-proximity measures such as the time to collision.

Automated Analysis and Validation of Right-Turn Merging Behavior

This article describes an automated approach for the analysis of right-turn merging behavior of vehicles. Traditional methods for collecting merging behavior data are labor intensive, suffer from reliability issues, are time consuming, and costly. Automated video merging behavior analysis is advocated as alternative data collection procedure resolving many shortcomings in the manual data collection. The main elements of the behavior analysis include merging conflicts, gap acceptance, and lane discipline. Traffic conflicts provide invaluable information that can be used to assess safety factors and to understand potential collision mechanisms. Gap acceptance is important for developing merging vehicles modeling frameworks. Lane discipline of merging vehicles is significant in showing potential aggressive and dangerous merging maneuvers and driver compliance to traffic rules. The article advocates automated computer vision as the engine to capture and analyze various merging behavior elements. The analysis is demonstrated using a case study from Doha, Qatar. A validation of the results was performed that demonstrated the soundness of the methodology and potential benefits for automated behavior data collection. The microscopic behavior data captured using the proposed automated methodology can be useful for use in road design, traffic management and safety evaluation.

Comparison of Time-Proximity and Evasive Action Conflict Measures

Interest has grown in using traffic conflicts for studying safety from a broader perspective than relying only on collision data. Traffic conflict analysis is typically performed through the calculation of traditional conflict severity measures such as time-to-collision and postencroachment time. These measures rely on road users getting within specific temporal and spatial proximity from each other and therefore assume that proximity is the surrogate for severity. However, this assumption may not be valid in some driving environments where close interactions between road users are common and sudden evasive actions are frequently used to avoid collisions. It is suggested that evasive action–based conflict indicators can assess the analysis in some less-organized traffic environments. This study focused on the severity evaluation of pedestrian conflicts. Pedestrian evasive actions were reflected mainly in variations of spatiotemporal gait parameters (step frequency and step length). The objective was to compare the use of time proximity and evasive action–based conflict indicators in evaluating the severity of pedestrian conflicts in different traffic environments. Video data from intersections in five major cities—Shanghai, China; New Delhi, India; New York City; Doha, Qatar; and Vancouver, British Columbia, Canada—were analyzed with automated computer vision techniques to extract pedestrian-involved conflicts and calculate conflict indicators. Results show that evasive action–based indicators were more effective in identifying and measuring the severity of pedestrian conflicts than time proximity measures in traffic environments such as Shanghai and New Delhi. However, evasive action measures did not show the same potential in Vancouver and Doha, where time proximity measures were more effective.

Automated Region-Based Vehicle Conflict Detection Using Computer Vision Techniques

An approach for vehicle conflict analysis based on three-dimensional (3-D) vehicle detection is presented. Techniques for quantitative conflict measurements often use a point trajectory representation for vehicles. More accurate conflict measurement can be facilitated with a region-based vehicle representation instead. This paper describes a computer vision approach for extracting vehicle trajectories from video sequences. The method relied on a fusion of background subtraction and feature-based tracking to provide a three-dimensional (3-D) cuboid representation of the vehicle. Standard conflict measures, including time to collision and postencroachment time, were computed with the use of the 3-D cuboid vehicle representations. The use of these conflict measures was demonstrated on a challenging data set of video footage. Results showed that the region-based representation could provide more precise calculation of traffic conflict indicators compared with approaches based on a point representation.

Models to evaluate the severity of pedestrian-vehicle conflicts in five cities

ABSTRACT Previous studies have shown that traditional traffic conflict indicators that depend on time-proximity are not a viable measure of conflicts severity in all driving cultures. Behavior-based indicators that are dependent on road-users evasive actions were shown to better reflect severity in less-organized traffic environments. The objective of this paper is to examine the use of time proximity-based and evasive action-based indicators on pedestrian conflicts in five major cities; Shanghai, New Delhi, New York, Doha, Vancouver. Time-to-collision is used as the primary time proximity indicator. Pedestrian evasive actions are reflected in the sudden variation of pedestrian gait parameters. Ordered-response models were utilized to relate both indicators to severity taking into account the unobserved heterogeneity in conflicts. Results show that the evasive action-based indicator is most effective in less-organized traffic environments such as Shanghai and New Delhi while the time proximity measure was shown effective in more structured environments such as Vancouver.

Before-after safety analysis using extreme value theory: A case of left-turn bay extension

There is growing interest in the use of traffic conflicts in before and after safety evaluations because of well-recognized quality and quantity problems associated with historical crash records. Most of these studies apply statistical techniques to compare the number of conflicts before and after the implementation of safety countermeasures. However, to identify the number of conflicts, a specific threshold for various conflict indicators needs to be used and the results of the evaluation can vary significantly depending on the selection of this threshold. As well, there is an issue with how to account for conflict severity in the evaluation. This study proposes adopting the extreme value theory approach to overcome these two issues. The approach was applied to a case of left-turn bay extension at three signalized intersections, and the automated traffic conflict technique was used to identify conflicts with TTC values from the video data collected from treatment sites and matching control sites. Generalized extreme value (GEV) models with different covariates were developed and compared. The results show that there are apparent shape change in the GEV distribution (i.e., from narrow peak up to high severities to wide spread with fewer conflicts at high severity levels) after the treatment, indicating reduction in conflict severity. The safety improvement is further confirmed by the total reduction of 63.9% in estimated crashes. Moreover, with the aid of GEV model, the most severe conflicts that are also rare and random are included into the OR calculation, and a significant reduction of 73.2% is found in the estimated most severe conflicts.

Evaluating the safety and operational impacts of left-turn bay extension at signalized intersections using automated video analysis

Left-turn lanes are commonly introduced to provide space to accommodate comfortable deceleration and adequate storage of turning vehicles. Operational shortcomings may arise due to inadequate length, including overflow and blockage of left-turn entrance by queues on an adjacent through lane. This study investigates the potential safety and operational benefits of treating left-turn lanes by extending the length further upstream a signalized intersection. Video data was collected at three treated left-turn lanes as well as three matched control lanes; all in both before and after treatment conditions. Safety parameters consisted of the counts and severities of traffic conflicts occurring on the left-turn lanes and inside the intersection. There was a marked reduction in traffic conflict counts in all treated sites. The overall treatment effect, which accounts for the simultaneous change in control sites, was 63.2% (p < 0.05). There was a marked reduction in frequency of traffic conflicts at different severity levels. The mobility benefit of the treatment was demonstrated in terms of the reduction in average travel time for left-turn as well as through vehicles. The count of traffic signal cycles with blocked left-turn entrance was considerably reduced after the treatment. The use of collision data gathered from more sites is suggested as potential future work to further evaluate this treatment.

Automated Pedestrians Data Collection Using Computer Vision

Active modes of travel such as walking are being encouraged in many cities to mitigate traffic congestion and to provide health and environmental benefits. However, the physical vulnerability of pedestrians may expose them to severe consequences when involved in traffic collisions. This paper presents three applications for automated video analysis of pedestrian behavior. The first is a methodology to detect distracted pedestrians on crosswalks using their gait parameters. The methodology utilizes recent findings in health science concerning the relationship between walking gait behavior and cognitive abilities. In the second application, a detection procedure for pedestrian violations is presented. In this procedure, spatial and temporal crossing violations are detected based on pattern matching. The third study addresses the problem of identifying pedestrian evasive actions. An effective method based on time series analysis of the walking profile is used to characterize the evasive actions. The results in the three applications show satisfactory accuracy. This research is beneficial for improving the design of pedestrian facilities to promote pedestrian safety and walkability.