Anti-Occlusion Scale-Adaptive Correlation Tracking

Target tracking at high-speed conditions is highly desirable for embedded systems, such as visual navigation. However, existing methods face challenges in achieving real-time processing due to algorithm complexity and computational limitations. To address this issue, we propose an anti-occlusion scale-adaptive correlation tracker (OSCT) based on a Kernelized Correlation Filter (KCF), which considers tracking confidence and multiple features in embedded systems. Specifically, we employ a discriminative correlation filter to estimate the scale of the tracking target. Subsequently, our tracker dynamically adjusts the learning rate by evaluating the confidence of tracking results. Moreover, the OSCT uses a re-detector to re-detect the target in case of tracking failure. Instead of relying on a single feature, our method integrates multiple features to address variation and occlusion in tracking objects. Experimental results demonstrate that: 1) OSCT achieves an average precision of 0.778 at 20 pixels, representing an improvement of over 8% compared to other high-speed trackers. 2) The average running speed of OSCT is 32 FPS, outperforming LCT (26 FPS). Therefore, our approach is well-suited for long-term tracking on embedded platforms.