LAFTrack: A Lightweight and Fast MOT Framework that Balances Appearance and Motion Features

With the advancement of computer vision technology, Multi-Object Tracking (MOT) has become a vibrant research field, holding significant value in various application scenarios such as video surveillance and autonomous driving. However, achieving tracking accuracy while maintaining real-time speed, especially in complex scenes, remains a challenge. This paper presents a lightweight and fast multi-object tracking (LAFTrack), which effectively integrates appearance features and motion features, enabling efficient real-time tracking in resource-constrained environments. The core of the LAFTrack framework lies in the Improved Spatial Similarity (ISS) module and the Spatio-Temporal Compensation (STC) module, which significantly enhance accuracy and robustness during the data association phase. Furthermore, this study employs the lightweight MobileNetV2 network as the foundation of the ReID module to reduce computational load while maintaining tracking accuracy. Experimental results on the MOT17 and MOT20 datasets demonstrate that LAFTrack maintains high precision while achieving a frame rate nearly double that of existing models, showcasing its advantage in real-time performance. Additionally, the lightweight design of LAFTrack makes it more suitable for deployment on edge devices, meeting the needs of time-critical tasks. The code is available at https://github.com/Leslie199909/LAFtrack.