Scale-adaptive vehicle tracking based on background information

Abstract

To solve the problem of low accuracy and poor robustness of vehicle tracking in complex traffic scenes, scale-adaptive vehicle tracking based on background information is therefore proposed to this paper. The traditional correlation filter tracking algorithm is less dependent on background information. This easily leads to tracking error. We propose to use the background information of the vehicle and the surrounding as a sample set to establish a position classifier. It transforms the target tracking problem into the classification of the target and the background. This also improves the position accuracy of the tracking target response point when the background is complex. The dimensions of the vehicle change as the relative distance between the vehicle and the camera changes, affecting the tracking reliability. This algorithm crops Histogram of Oriented Gradient (HOG) features of the different-scale vehicle images and establishes a scale classifier. It determines the best scale of the target built on the output response peak of the scale classifier. This improves the adaptability of classifier against vehicle scale change. Extensive experimental results demonstrate that the method improves the accuracy and robustness of vehicle tracking significantly.