Research Depth-information Based Of Kernelized Correlation Filters Tracking Algorithm

Visual object tracking is a fundamental task in computer vision field. In order to deal with all kinds of complex tracking scenario, achieve good tracking effect, now more and more tracking algorithm combining tracking and target detection. The depth information not only can effectively deal with the problem of large area of the target, but also because of the acquisition process has become more and more simple, making it become more important in the field of visual tracking. Most tracking algorithms now use depth information as supplementary information to judge the occlusion, and do not make full use of the depth of the role of information. In recent years, the kernelized correlation filters discriminant tracking algorithm performance is particularly prominent. However, when the target shaded with large scale and the more serious the target deformation, the performance of the algorithm is greatly reduced. Therefore, the Depth-information based of Kernelized Correlation Filters Research Algorithm Tracking(DKCF) is proposed.The effect of depth information on the tracking algorithm is analyzed. Most of the current tracking algorithm is extracting feature information in the two-dimensional image, does not indicate the actual position of the target in the scene, the lack of spatial information object. The relative position of the target can be extracted accurately, which can be used to distinguish the target and not target. The current frame of the KCF maximum response point corresponding depth map of the corresponding position, this position is calculated depth map search area confidence map. Then the Gaussian sampling method, the probability of each sample point target area size and selection probability and maximum points from the current frame KCF maximum response point of fusion, the end frame of the tracking target coordinate calculation. Through the continuous update of the target zone depth value and KCF template, the proposed algorithm not only has the KCF algorithm to construct a stable appearance model and target template can better against occlusion. Experiments show that DKCF algorithm can effectively improve the accuracy of tracking and can deal with the problem of large area occlusion and large deformation...