• Ying Ni, Menglong Wang, Jian Sun, and Keping Li.
• Department of Traffic Engineering, Tongji University, Shanghai 201804, China. Electronic address: ying_ni@tongji.edu.cn.
• Accid Anal Prev. 2016 Nov 1; 96: 118-129.

AbstractPedestrians are the most vulnerable road users, and pedestrian safety has become a major research focus in recent years. Regarding the quality and quantity issues with collision data, conflict analysis using surrogate safety measures has become a useful method to study pedestrian safety. However, given the inequality between pedestrians and vehicles in encounters and the multiple interactions between pedestrians and vehicles, it is insufficient to simply use the same indicator(s) or the same way to aggregate indicators for all conditions. In addition, behavioral factors cannot be neglected. To better use information extracted from trajectories for safety evaluation and pay more attention on effects of behavioral factors, this paper develops a more sophisticated framework for pedestrian conflict analysis that takes pedestrian-vehicle interactions into consideration. A concept of three interaction patterns has been proposed for the first time, namely "hard interaction," "no interaction," and "soft-interaction." Interactions have been categorized under one of these patterns by analyzing profiles of speed and conflict indicators during the whole interactive processes. In this paper, a support vector machine (SVM) approach has been adopted to classify severity levels for a dataset including 1144 events extracted from three intersections in Shanghai, China, followed by an analysis of variable importance. The results revealed that different conflict indicators have different contributions to indicating the severity level under various interaction patterns. Therefore, it is recommended either to use specific conflict indicators or to use weighted indicator aggregation for each interaction pattern when evaluating pedestrian safety. The implementation has been carried out at the fourth crosswalk, and the results indicate that the proposed method can achieve a higher accuracy and better robustness than conventional methods. Furthermore, the method is helpful for better understanding underlying levels of safety from the behavioral perspective, which can also provide evidence for targeted traffic education on proper behaviors. Copyright © 2016 Elsevier Ltd. All rights reserved.

21 keywords...

hide…