Performance Analysis of Re-ID in YOLOv9 and BoT-SORT Algorithm for Multi-object Detection and Tracking

This paper proposes a method for integrating YOLOv9 and BoT-SORT algorithms for multi-object tracking in a multi- camera environment. Particularly, it focuses on analyzing the impact of re-identification (Re-ID) techniques on the overall tracking system performance. Three Re-ID methods—MobileNetV2, OSNet, and OSNet-AIN—are evaluated using the MSMT17 dataset and then applied to the BoT-SORT algorithm to compare multi-object tracking performances. Experimental results show that BoT-SORT with OSNet-AIN achieves the best performance in HOTA, MOTA, and IDF1 metrics. This demonstrates that effective Re-ID techniques can significantly contribute to improving object tracking performance in complex real-world environments.