An effective video analysis method for detecting red light runners

This paper presents a novel method for automatic red light runner detection on a video, which is fundamentally different from the concept of conventional red light camera systems. In principle, it extracts the state of the traffic lights and vehicle motions without any physical or electronic interconnections to the traffic light control system or the buried loop detectors. Purely from the video, the new method first constructs a traffic light sequence and then it estimates vehicle motions beyond the stop line while the light is red. In the former, the spatial and temporal relationships of individual traffic lights are utilized. In the latter, the concept of virtual loop detector has been introduced to emulate the physical loop detectors. A prototype was implemented based on this method and was tested in a number of field trials. The results show that the new method is able to detect multiple red light runners in multiple lanes. It is also capable of tolerating a number of hostile but realistic situations such as: 1) minimum number of traffic light; 2) pseudomotions due to shadows; 3) poor contrast; 4) pedestrian motions; and 5) turning vehicles