Automated Detection of Spatial Traffic Violations through use of Video Sensors

It is plausible that road use behavior that leads to traffic violations can be a contributing factor to the failure mechanism that prompts a collision. The detection and the understanding of traffic violations, therefore, can be key components of a sound safety diagnosis. Recent advances in computer vision techniques have enabled the conduct of automated and large-scale analyses of various approaches to safety diagnoses, including traffic conflicts and violation analysis. This paper proposes and compares two approaches to detect vehicular spatial violations. The approaches are (a) k-means clustering and (b) pattern matching with use of the longest common subsequence (LCSS) similarity measure. The purpose of this study was to learn what constituted normal movement patterns and to interpret any discrepancy between these patterns and observed tracks as an indication of a traffic violation. Each of the approaches was applied to detect U-turn violations on an urban intersection in Kuwait City, Kuwait. Violation detection on the basis of LCSS generally was superior to detection with piecewise k-means clustering, especially when low false detection rates were desirable. This expected differential was the result of the performance of LCSS matching on observed road user positions, and not on an approximated model.