Research on Vehicle Target Tracking in Satellite Videos Based on an Improved Kernelized Correlation Filter Algorithm

Low Earth Orbit satellites provide continuous monitoring of target areas on the Earth’s surface, capturing dynamic and rich remote sensing data through video imaging. However, vehicles in satellite video are typically small in size, and the background often contains significant interference and redundant information. Additionally, compared to conventional vehicle tracking tasks, targets in satellite video are more susceptible to occlusion. To tackle these challenges, this paper introduces an improved kernel correlation filter algorithm for vehicle tracking in satellite video. The proposed algorithm not only achieves high-precision tracking but also effectively recovers the tracked target in cases of target loss. Initially, the Kalman filter is introduced to compensate for the target’s motion trajectory, enhancing the overall tracking performance. Furthermore, to address issues related to target loss and occlusion, a re-detection strategy is proposed, which restores target tracking through pyramid search and sub-pixel matching techniques when target loss is detected. The experimental results show that the proposed algorithm outperforms the second-best ASLA by 7.1% in tracking precision and exceeds the second-best ECO by 8.7% in tracking success rate, highlighting its potential and advantages in practical applications.