Moving objects tracking from most probable regions and eliminating camera motion

This paper presents a novel method for moving object tracking in different scales. There are researches in tracking objects but most of them focus on specific subject and fail in some conditions such as changing position, moving camera, changing scale because of the distance variations. Camera movement is one of the most challenging events which causes to have a lot of fake moving objects in scenes. In this paper we modify KLT (Kanade- Lucas- Tomasi) algorithm by spectral residual in different Gaussian pyramid scales and extract positions with high probability of objects presence. To achieve perfect tracking, consecutive frames are rectified by finding the best matches between features points and remove undesired effects of camera movements. To evaluate the proposed approach, we arrange experiments using standard databases and compare with the other methods reported in the literature. The results indicate that the proposed approach is capable of detecting and tracking all the moving objects in acceptable accuracy rate, i.e., over 90% accuracy in average in all challenging databases.