Multi-Object Tracking Method for Aerial Images Based on ReID-Byte

Aiming at the problems of ID Switch and track fragment caused by the ground target occlusion in aerial video tracking, a multi-object tracking(MOT) method based on ReID-Byte is proposed. This method introduces the ReID model on the basis of the Byte algorithm to extract the appearance feature of the targets. This paper uses the multi-task learning method to train the YOLOX detector and the ReID model, and the detection confidence is used as the measure to separate the detection results to the high-scoring boxes and the low-scoring boxes for inter-frame target association processing. The appearance-motion model is used to match the trajectory with the high-scoring targets, and the motion model is used to match the trajectory with the low-scoring targets. The test results of the UAV aerial dataset show that the method proposed in this paper improves the tracking accuracy and can effectively handle the ID Switch and track fragment problems caused by the target occlusion.