Study Of Nighttime Video Vehicle Detection Under Complex Environment And Luminance Transformation Model Between Day And Night For Traffic Video

Intelligent Transportation System is an effective modern traffic management method, which is based on traffic information collection system. Compared to traditional traffic information collection systems, video detection systems are more flexible and have multi-additive-functions. Therefore, video detection systems are preferred. However, since video detection systems are very sensible to roadway environment and the quality of video, their detection accuracies are not high enough and their detection algorithms need to be improved. Practical applications have indicated that, these systems have even lower accuracies at night, especially at rainy night or on HOV line. Most previous researches on nighttime video vehicle detection studied vehicles'headlights, and they simply assumed that headlights have regular shapes in video images, but had not taken the following complex situations into consideration adequately: conjunctions among headlights, fog lights, indictors and decorative lights, presences of boundary lights, inverted reflections of those lights in road seeper.Therefore, this paper focuses on nighttime situation instead of treating all-weather video as a whole, and studies it separately, aiming at complicated situations such as rainy weather, presence of large vehicles and/or traffic jam. After analyzing diverse attributes of video on different roadway conditions and various representations of the lights in videos, a headlight extraction, pairing and tracking algorithm (HLEPT) is presented. The algorithm extracts headlights by binarization and morphological methods. Based on camera calibration, the algorithm finds the projection coordinates of headlights on the roadway plane subject to camera lens. According to physical construction and movement regulation of the vehicle, a series of regulations of headlight pairing, grouping by individual vehicle and...