An improved thinning algorithm for human body recognition

Abstract

This paper introduces a robust and reliable method of human body recognition for visual surveillance systems. The method is developed based on the skeleton of moving object, which can be got by an improved thinning algorithm. In the paper we first describe some thinning algorithms for binary images, including OPTA thinning algorithm, Zhang's thinning algorithm and Rosenfeld's thinning algorithm. Comparing the performance of these thinning algorithms, we found that all the described algorithms have their own disadvantages. For instance, OPTA thinning algorithm produces spurious branches and noise, which may result in misidentification of the pattern, Zhang's and Rosenfeld's thinning algorithm are all time consuming and don't satisfy the one-pixel-width requirement. So we propose an improved thinning algorithm, which is much faster and the skeleton satisfies one-pixel-width requirement. In addition, a human legs-detecting algorithm based on the skeleton of the moving object is presented. Experimental results of the proposed thinning algorithm and the human legs-detecting algorithm on a variety of human images will be shown.