Research On Nighttime Vehicle Detection Technology Based On Video

With the development of economic and society,highway traffic has also been a great deal of development.Intelligent Transport System(ITS)was proposed in the 1960s.It has played an effective mitigation effect on the increasingly tense transportation situation.Vehicle detection is a key link in Intelligent Transportation System.Video-based vehicle detection is monitored using a camera.When the vehicle enters the monitoring area,the image of the picture changes,and the presence of the vehicle can be detected by these changes.And the required real-time traffic information and monitoring parameters can be obtained.Video-based vehicle detection is the basic link in the whole intelligent transportation system,which is of great significance to the subsequent analysis of vehicle behavior.The main work of this thesis is:(1)This thesis proposes a nighttime vehicle detection algorithm based on the characteristics of rear lights.First,using Hough transform to extract the lane,set up a virtual coil adapted to the lane.Then,the area where the tail lights may exist is divided according to the color threshold.And use the OTSU algorithm to automatically determine the threshold to separate the rear lights target.Finally,pair and track the rear lights in the area of virtual coil,and complete the whole process of traffic statistics.The algorithm is tested by the actual road video collected in Xiamen,and in the Matlab R2016a environment for testing.Compared with the existing headlight-based method,this method can effectively calculate the traffic statistics at night,and the detection accuracy is over 95%.(2)This thesis improves a nighttime vehicle detection and tracking algorithm based on AdaBoost classifier.First,the cascade classifier based on the Hard Negative Sample classifies the video images and obtains the hypothesis windows of the rears of the vehicles.The design of the two-layer classifier increases the training time but reduces the detection time of the classifier,which improves the performance of the algorithm as a whole.And then according to the color and motion information of the hypothesis windows,filter the hypothesis windows of the rears of the vehicles.Finally,Kalman filter is used to track the vehicle and complete the whole process of the algorithm.Compared with the normal AdaBoost classifier,the error rate of the improved method is less than 20%,which greatly reduces the false positive rate of the classifier.Both of these two algorithms in this thesis can achieve high detection accuracy,and are effective nighttime vehicle detection methods.The future will further enhance the robustness and real-time of these algorithms,and will try to apply them into vehicle identification.