Object Tracking in Satellite Videos Based on Improved Kernel Correlation Filter Assisted by Road Information

Video satellites can stare at target areas on the Earth’s surface to obtain high-temporal-resolution remote sensing videos, which make it possible track objects in satellite videos. However, should be noted that object size videos is usually small and has less textural property, moving are easily occluded, puts forward higher requirements for tracker. In order solve above problems, consider image contains rich road information, used constrain trajectory of a video, this paper proposes an improved Kernel Correlation Filter (KCF) assisted by information objects, especially when occluded. Specifically, contributions as follows: First, tracking confidence module reconstructed, integrates peak response average correlation energy map more accurately judge whether Then, adaptive Kalman filter designed adaptively adjust parameters according motion state object, improves robustness reduces drift after Last but not least, strategy recommended, searches with constraints, locate accurately. After improvements, compared KCF tracker, our method precision 35.9% success rate 18.1% speed 300 frames per second, meets real-time requirements.