VEHICLE TRACKING IN SATELLITE VIDEOS BASED ON TEMPORAL CHANGE INFORMATION

Vehicle tracking is a crucial task in the intelligent interpretation of remote sensing (RS) videos. However, from the perspective of satellites, the scale of vehicle is quite small and information directly extracted from satellite videos is highly susceptible to interference. This makes it difficult to accurately track vehicles, significantly reducing the precision of the tracking. To address these challenges, TCINet, a tracking method that uses temporal change information for interference suppression and change prediction, is proposed in this paper. Firstly, the temporal change information extraction module (TCIE) is proposed to suppress temporal change information unrelated to vehicles, thereby indirectly improving the model’s capacity to extract information pertinent to vehicles. Based on this, the temporal change information prediction module (TCIP) is proposed to predict the future temporal change information of the vehicles, and the predicted information is then integrated with the information from the current frame to optimize and supplement vehicle information in the subsequent frame. Experimental results on the SatVideoDT dataset validate the effectiveness and superiority of our proposed method.