A Video-moving Object Tracking Algorithm Research Based On Weighted Correction And Bilateral Filtering

In computer vision, target tracking is always one of the most difficult technologies and it has been applied in various fields. The traditional tracking algorithm can cause accumulated error because of outside interference, then it make tracking template drift in the process of continuous track and affect the target tracking accuracy. How to suppress drift and correcting drift effectively, to realize high precision and real time target tracking have become the most written topics in computer vision.In this paper, we mainly analyze the template drift phenomena in video target tracking.According to the results of drift phenomenon analysis, we proposed a target track algorithm based on additive synthesis drift correction and a target track algorithm based on additive weighted drift correction. According to the actual application demand, we have a further research on the multiple targets tracking algorithm while it’s in emerged-occlusion condition.And provide a multi-objective track algorithm base on corner detection with bilateral filtering. The main contents of the paper:(1) Analyze the reason of drift phenomena in traditional video tracking, study the research status about suppress drift, correcting drift and its development trend.(2) We mainly study traditional Lucas- Kanade algorithm, additive synthesis Affine Image Alignment algorithm and passive drift correction algorithm, and explain the reason of the drift, and provide a new improved algorithm in the end: a target track algorithm based on additive synthesis drift correction. The results show that the new algorithm can suppress drift and correcting drift.(3) According to a target track algorithm based on additive synthesis drift correction, we designing a new weighted coefficient optimization method, by introduce a new weight coefficient in the algorithm, we proposed a target track algorithm based on additive weighted drift correction, it solved the template drift in tracking. The results show that the new algorithm has a more effective calculation and better robustness.(4) We proposed an improved corner detection method based on bilateral filtering, by replaced Gaussian low-pass filter with the bilateral filter, and using double threshold corner detection to extract the relevant characteristics of angular point to improve the stability of new methods. Use K neighbor classifier to classify the relative angular point information in the tracking area, it can separate the tracking target. The results show that the new algorithm can track the occlusion multi-target accurately.