Development of a Human Detection and Tracking Using Multiple-Kernels

In this paper, we propose an innovative human tracking algorithm in a mass of videos recorded from cameras placed in public places. In a visual surveillance system with static cameras, target often fails when part of the object has similar color with the background or when the target is occluded, therefore we use multiple kernels with the possibility of selection of 2 and 4 kernels for each person in the video frame. In order to provide more robust tracking performance, histogram of the target should be changed per frame. Total cost function and an adaptively changeable weights assigned to each kernel based on the similarity between the histogram, the exact location of the target image based on mean shift approach will be achieved. Experimental results have shown the favorable performance of the proposed algorithm, which can successfully track human tracking in public places.

کلیدواژه ها:

Human tracking ، mean shift ، multiple kernels ، visual surveillance