Robust visual tracking under light spot interference

作者：Siyuan Wang, Haotong Ma, Chen Wang, Han Luo, Yu Jiang · 发表于：International Conference on Machine Learning and Large Models · 年份：2026 · DOI：10.1117/12.3117277 · 研究领域：Engineering

Mainstream object tracking models primarily focus on exploring diverse attention mechanisms, advancing multi-camera perspective-based designs, and developing multi-modal frameworks. In contrast, the most remarkable progress has been witnessed in the fields of image generation and image restoration—achievements that have not yet been cross-domain integrated into object tracking models. To tackle the challenge in engineering practice where light beam echo spots degrade tracking performance, we propose a novel tracking framework that fuses image restoration with object tracking. Specifically, the framework first restores regions occluded by light spots, followed by performing tracking on the restored images. This design significantly enhances the performance of object tracking models on datasets contaminated by light spot interference.