Research On Kenerl-based Object Tracking Algorithms

Computer Vision is a hot research topic in the field of machine vision and artificial intelligence.Video object tracking technology is an important research area and the theoretical foundation in thefield of computer vision. In recent years, object tracking is applied in many fieldes such as medcial,surveillance, perceptual user interfaces. Among the various tracking algorithms, the kenerl-basedobject tracking algorithm has become popular due to its simplicity, efficiency and goodperformance.Chapter I of this paper briefly introduces the research background of the object tracking andtracking algorithm. Chapter II studies kenerl-based tracking algorithm, the algorithm calculate thecolor histogram according to the color features of the target and candidate target kenerl probabilitydensity estimation, then calculate the similarity degree between them through the similarity function,to obtain the maximum similarly candidate model which is positioned as the tracking target in thecurrent frame. The experimental results show that the method has good performance in the trackingprocess, but when the tracking target complex movement or target model similar to the backgroundcolor, the tracking performance will drop.To improve the uniqueness of target feature representation in the traditional kenerl-based objecttracking algorithm, Chapter III presents kenerl-based object tracking to increase the texture featurein the target and candidate target model using the joint colortexture histogram. The experimentalresults show that the proposed method can not only improve the tracking accuracy, but alsorobustly track the target under complex movements. However, it increases the complexity of thealgorithm to real-time tracking.In order to improve the tracking performance in the situation that the target is similar tobackground and to reduce the calculation burden of the algorithm in the third chapter, Chapter IVproposes the modified background-weighted algorithm, which only weights the target model but notthe target candidate model. The experiment results show that the algorithm alleviates effect of thetarget background and complex montion on tracking performance, and is not sensitive to the impactof illumination.