<![CDATA[Multi-object tracking (MOT) is a fundamental task in computer vision that underpins applications such as intelligent surveillance, autonomous driving, and crowd analysis. The primary challenge in MOT lies in maintaining identity consistency under frequent occlusions while ensuring real-time performance on resource-constrained devices. This study proposes LITE-BoostTrack, a hybrid tracking framework that combines the confidence-based association mechanism of BoostTrack with the lightweight embedding strategy of the Lightweight Integrated Tracking and Embedding (LITE) architecture. The proposed model extracts appearance descriptors directly from the internal feature maps of the YOLOv8 detector, thereby eliminating the need for an external re-identification network. This design significantly reduces computational complexity while preserving reliable identity association. Experiments were conducted on the MOT20 benchmark using standard MOT evaluation metrics, including HOTA, MOTA, IDF1, IDSW, and FPS, to assess both tracking accuracy and runtime efficiency. The results show that LITE-BoostTrack achieves a HOTA of 27.31 and IDF1 of 37.48, outperforming LITE-BoT-SORT (HOTA 25.73, IDF1 33.88), while reducing identity switches by 37% (2,939 vs. 4,674) and maintaining real-time performance at 13.22 FPS. These outcomes demonstrate that substantial efficiency gains can be achieved through detector-level feature integration without introducing additional deep embedding modules. Although occasional failures still occur under severe occlusion, LITE-BoostTrack provides a balanced and practical solution that effectively combines accuracy and efficiency for real-time multi-object tracking in edge-computing and embedded vision systems.]]>
