Construction and Research of Multi-Target Tracking Method in Low Light Scene Based on PP-YOLOE+m and PP-Tracking

Multi-target tracking in low-light scenes is an important research direction under of computer vision. Effective multi-target tracking methods in low-light have scenes are of great significance for applications such as video surveillance, unmanned driving, and intelligent security. Under this environment, due to insufficient light and noise interference, it faces problems such as unstable low-light image enhancement effect, target association accuracy and robustness, algorithm real-time performance and computational efficiency. Under this paper, the PP-YOLOE+m model is used for multi-target tracking in low-light scenes. By training and optimizing the PP-YOLOE+m model, the target detection ability of the PP-YOLOE+m model in low-light scenes is improved. At the same time, combined with the PP-Tracking algorithm, the accurate tracking and ID maintenance of the target can be realized. Using the ExDark dataset to evaluate the effect of the multi-target tracking method in this paper, the accuracy of multi-target tracking has been improved by 3.6%, which verifies the effectiveness and robustness of the method in this paper. Experimental results show that, the method proposed can more accurately track multiple targets, and provides guidance for vision applications in low-light scenes in low-light environments.