Research on Continuous Object Real-time Tracking Based on SIFT and Particle Filter

Abstract

Object track is a hot topic in the field of Cyber-Physical Systems (CPS). Because the existing object tracking algorithm, which is based on Scale-invariant feature Transform (SIFT) and Particle Filter (PF), will make the feature points of target model be all deleted and not be added during the period of the full occlusion of the target, it isn’t able to work. According to this problem, this paper introduced adaptive updating target model and object matching. The specific method is taking a object tracking method of SIFT and PF, taking Random sample consensus (RANSAC) to exclude error matching, stopping updating object model while establishing candidate one due to the losing of tracking object, and matching objects by Best Bin First (BBF) optimized by k-dimensional tree (k-d tree). The simulation results show that this method was robustness when the object reappeared after full occlusion.