Research And Implementation Of Pedestrian Detection And Tracking Based On Vehicle-loaded Video

The technique of pedestrian detection and tracking in urban traffic scenes is one of the key techniques for Intelligent Vehicle Navigation System. And the research on it will do great contribution to the development of Intelligent Vehicles and the safety of city traffic. The occurrence of traffic accidents brings human society great harm. So, how to reduce the occurrence of traffic accidents, detect and track pedestrians in front of vehicles, and then take measures to protect pedestrians immediately, avoid pedestrian collision are the research topic of Intelligent Vehicle Safety Driving Assist System.In the pedestrian detection process, the method based on the AdaBoost classifier and HOG features is presented in this paper. Firstly, pedestrian images are segmented to get the candidate pedestrian areas. Secondly, pedestrian samples are selected, and HOG features of the samples are extracted, which are used to training AdaBoost classifier. Finally, the AdaBoost classifier trained is used to recognize the candidate pedestrian areas.In the pedestrian tracking process, the traditional Mean Shift tracking algorithm is used in color image sequences, and the improved Mean Shift tracking algorithm is presented in infrared image sequences, which uses histogram of gradient’s direction as the target feature, constitutes the core of the Mean Shift tracking procedure to reduce the search regions of pattern matching with the finite iteratives to achieve the quick convergence. Infrared images are the gray images which have not color informations and contain the little texture details. The traditional Mean Shift tracking algorithm, using the color histogram as the target feature, is only applied in color image sequences. The problem that the traditional Mean Shift tracking algorithm can not be applied in infrared image sequences is solved by the improved Mean Shift tracking algorithm.To improve the accuracy of moving pedestrian tracking in the car video, the combination algorithm based on the Mean Shift algorithm and the kalman filter is presented in this paper. Kalman filter is used to forecast the possible position of target in the current frame, and then the real position is searched near the possible position by the Mean Shift algorithm. The experiments show that this method has good tracking results.