Object Tracking in Satellite Videos With Distractor–Occlusion-Aware Correlation Particle Filters
Yangfan Li,Nan Wang,Wei Li,Xiong Li,Mengbin Rao
DOI: https://doi.org/10.1109/tgrs.2024.3353298
IF: 8.2
2024-02-02
IEEE Transactions on Geoscience and Remote Sensing
Abstract:With the advancement of high-resolution remote sensing satellites, the tracking of high-value targets such as planes and ships within satellite videos has become imperative. In recent years, several object tracking methods designed for satellite videos based on correlation filters have been proposed. However, these traditional correlation filters typically identify the location with the highest response value on the response map as the target position. In the context of satellite videos, where targets are often very small and surrounded by numerous similar objects, depending only on response values to determine the target's location can easily lead to interference from nearby objects, resulting in tracking failures. Moreover, targets frequently encounter occlusion during their motion, further complicating tracking tasks due to the absence of distinctive target appearance features and leading to the issue of model drift. To address these challenges, we propose a novel distractor–occlusion-aware correlation particle filters. Instead of determining the position with the maximum response value, our method initially selects the top response values from the response map, creating a pool of candidates. Subsequently, we introduce an innovative quality score, rooted in motion information and response scores related to the target, for each candidate. Finally, these quality scores are employed to filter the most suitable candidate. This novel distractor-aware module effectively equips our tracking method to perform well in the presence of distractors. Additionally, to handle occlusion, we integrate the occlusion-aware module into the correlation particle filters, improving the tracker's performance in occluded scenarios. To ensure the effective collaboration of the distractor-aware module and the occlusion-aware module, we introduce the dual Kalman Filter method. Our comprehensive experiments conducted on the SatSOT datasets conclusively demonstrate the effectiveness and superiority of our proposed tracking method. The code will be available at https://github.com/liyangfan0/DOCPF.
imaging science & photographic technology,remote sensing,engineering, electrical & electronic,geochemistry & geophysics