Research On The Algorithms Of Moving Object Tracking In Video

Video target tracking technology has been widely used in many fields of military weapons, traffic monitoring and security. This paper mainly focuses on the problem that the target is occluded or the background has similar color with the target which may lead to error tracking. Particle filter is a good filtering algorithm developed in recent years, which is widely used in target tracking. This paper mainly research on the particle filter algorithm and make some improvements in the aspects of target feature extraction, target model and model update algorithm.In the feature extraction stage, the two features are both used to describe the target together, which means we extract the color histogram feature and CBLID feature at the same time. CBLID is a new type of target feature description algorithm, which is inspired by two theories of the classic SIFT algorithm and the DAISY descriptor. In a word, we accelerate the feature extraction process.In target tracking stage, we need to build and update two target models, which are color target model and CBLID model. Color model’s update depends on the number of matched CBLID feature points between CBLID model and candidate fields. We update both two models only when the matching number of CBLID feature points is larger than the threshold value, or only update the CBLID model. The two features will be given a weight each in the particle weight updating period, the weight of each feature describes the importance of them, and the weight is adaptive related to the environment.After comprehensive tracking experiments of the proposed algorithm, we find that our proposed algorithm has better robustness and higher tracking accuracy, and it is faster than that of particle filter algorithm based on SIFT.