Multi-Object Tracking Method Based on Improved YOLOv11

Multi-object tracking (MOT) has significant academic value and application potential in scenarios such as video surveillance, autonomous driving, security deployment, and motion behavior analysis. The challenges it faces primarily include common issues such as target occlusion, scale variation, and appearance similarity interference, as well as the persistence of multi-object cross-frame identity matching. To address these challenges, this paper proposes a multi-object tracking algorithm based on an improved YOLOv11. The method enhances YOLOv11's C3K2 module by introducing the MetaFormer module and replaces its neck structure with RepGFPN. Additionally, the BoostTrack++ algorithm is incorporated to further improve the performance and stability of multi-object tracking. Experiments conducted on the public datasets CrowdHuman and MOT17 show that the proposed method achieves significant improvements in both object detection and tracking performance, validating its effectiveness.