Research on Multi-object Tracking Algorithm Based on Person Recognition and Detection Fusion

To solve the problem of poor tracking performance in complex scenes such as scale change and frequent occlusion, a multi object tracking method based on the fusion of person detection and person recognition is proposed. Firstly, the improved DLA-34 network is used as the backbone network, and the residual block is improved to improve the object detection accuracy; Then a Re-ID module based on time graph convolution neural network is designed to effectively solve the problem of person identity switching and tracking object loss caused by occlusion and small object in video. Finally, the effectiveness of the algorithm is verified on MOT16 and MOT17 datasets, and the accuracy and tracking accuracy of multi- object tracking are improved. The experimental results show that the network has good detection and tracking effects.