Infrared Target Tracking Algorithm Based on Motion Features and Contour Features

With the rapid development of target detection technology, MOT(multi-target tracking) technology has also been developed rapidly. For the field of infrared surveillance, the background is relatively complex and appearance features such as color are lost, so the tracking of infrared targets has been a challenging topic. In order to improve the speed and accuracy of IR target tracking, this paper explores a tracking method based on IR target motion features and contour features. The method follows the paradigm of TBD (Tracking by Detection) in multi-target tracking. Target detection uses a modified yolov5x as a detector, a Kalman filter for target motion state estimation, HOG(histogram of oriented gradients) features of the target frame map for re-identification, and dynamic threshold matching is used. At the same time, the low confidence detection frame is also involved in matching for potential occlusion problems. The experimental results show that the tracking algorithm has better results in IR(infrared radiation) complex permutation environment. Compared with DeepSORT algorithm, The speed of the tracking is improved by 40.2% and the accuracy is improved by 2.3%.