An improved vehicle-pedestrian near-crash identification method with a roadside LiDAR sensor

• Published in: Journal of safety research 2020-06, Vol.73, p.211-224
• Main Authors: Wu, Jianqing, Xu, Hao, Zhang, Yongsheng, Sun, Renjuan
• Format: Article
• Language: English
• Subjects: Accident Prevention - instrumentation; Accident Prevention - methods; Accidents, Traffic - prevention & control; Accidents, Traffic - statistics & numerical data; Algorithms; Automobile Driving - statistics & numerical data; Classification; Clustering; Data Collection; Data processing; Encroachment; Humans; Identification; Lidar; Near-crash identification; Pedestrian safety; Pedestrians; Pedestrians - statistics & numerical data; Roadside LiDAR; Roadsides; Safety; Traffic accidents & safety
• ISSN: 0022-4375; 1879-1247
• DOI: 10.1016/j.jsr.2020.03.006

•A novel method for pedestrian-vehicle near-crash identification is developed.•The roadside LiDAR sensor is used to collect the near-crash events.•Three indicators: PET, PSD, and CPI are applied for the risk classification. Problem: Potential conflicts between pedestrians and vehicles represent a challenge to pedestrian safety. Near-crash is used as a surrogate metric for pedestrian safety evaluations when historical vehicle–pedestrian crash data are not available. One challenge of using near-crash data for pedestrian safety evaluation is the identification of near-crash events. Method: This paper introduces a novel method for pedestrian-vehicle near-crash identification that uses a roadside LiDAR sensor. The trajectory of each road user can be extracted from roadside LiDAR data via several data processing algorithms: background filtering, lane identification, object clustering, object classification, and object tracking. Three indicators, namely, the post encroachment time (PET), the proportion of the stopping distance (PSD), and the crash potential index (CPI) are applied for conflict risk classification. Results: The performance of the developed method was evaluated with field-collected data at four sites in Reno, Nevada, United States. The results of case studies demonstrate that pedestrian-vehicle near-crash events could be identified successfully via the proposed method. Practical applications: The proposed method is especially suitable for pedestrian-vehicle near-crash identification at individual sites. The extracted near-crash events can serve as supplementary material to naturalistic driving study (NDS) data for safety evaluation.