Vehicle Detection and Tracking in Remote Sensing Satellite Vidio based on Dynamic Association

Abstract

Since remote sensing video satellites can continuously observe a certain target area and obtain multitemporal remote sensing images, it makes the surveillance of thousands of moving objects on the wide area possible. Vehicles are a kind of important and typical objects for remote sensing detection and tracking. In the paper, we propose an efficient method to detect and track vehicles in multi-temporal remote sensing images including two stages: Vehicle detection stage and tracking stage. In the vehicle detection stage, we use background subtraction and combine road prior information to improve accuracy and efficiency and reduce search space. In the tracking stage, we improve the traditional association matching method, which apply more dynamic association methods and more practical state judgment rule. In addition, we divide tracking objects into groups to further improve the accuracy. Our method is evaluated on remote sensing video dataset. According to experiment result, the proposed method can detect and tracking vehicle objects and correct the misdirected objects by the dynamic association structure. In the stable tracking stage, tracking quality is 96%. The experimental results show effectiveness and robustness of the proposed method in detection and tracking of vehicle objects from multi-temporal remote sensing images.