Multi-Target Tracking Algorithm Based on Improved ByteTrack

Aiming at the problems of target loss, slow speed and easy to be affected by noise in multi-target tracking algorithm, a multi-target tracking algorithm with improved ByteTrack algorithm is proposed. In terms of detectors, YOLOv8 is used to replace YOLOX, the number of C2f modules in Backbone structure is increased, and FasterNet modules and EMA attention mechanism are introduced to improve precision and reduce the number of parameters. CSPStage module of Neck part is redesigned to optimize the exchange of high-level semantic and low-level spatial information. The Wise-IoU loss function, which replaces CIoU, decreases the development of hazardous gradients. In the aspect of motion model, the noise adaptive Kalman filter is proposed, and the detection confidence is used to correct the noise and its prior probability, thus improve the tracking accuracy and stability. Experiments show that compared with the baseline algorithm, the detection indexes mAP0.5 and MAP0.5:0.95 of the improved algorithm are improved by 2.0% and 1.4% in MOT17 data set, and the tracking indexes MOTA and HOTA are improved by 0.64 and 2.74 in MOT20 data set, respectively. The indexes of the improved algorithm are all improved, which proves the effectiveness of the proposed method in improving the accuracy and stability of target detection and tracking.