Multi-Object Tracking in Crowded Scenes by Integrating Image Motion Compensation Mechanism and Improved ByteTrack Algorithm

Abstract

This study proposes a multi-object tracking method in dense crowd scenes that integrates the image motion compensation mechanism and improved ByteTrack algorithm. This method aims to improve detection accuracy, enhance the model's perception of smallscale targets and occluded targets, and significantly alleviate the performance bottleneck of multi-object tracking in dense scenes. Subsequently, it performs secondary matching on the low-scoring frames corrected by camera offset compensation and introduces Kalman filtering with adaptive noise adjustment to improve the robustness of trajectory prediction. Finally, through the comprehensive evaluation module, the tracking indicators are output. The results show that the F1 score in different scene complexities is overall greater than 85%. When the intersection ratio threshold is 0.5, the average accuracy is 95.2%. In practical application tests, the average recall rate of the research method exceeds 85% across different detection numbers for each picture. Its overall processing delay time under different population densities is less than 100 ms, and its overall missed detection rate in different target size intervals is less than 10%. Experimental results indicate that the proposed method exhibits competent small target detection capability and provides stable visual analysis under the tested conditions.