Target Detection Algorithm Based on Unconstrained Environment for Video Human Motion

Abstract

The existing human target detection algorithm can only detect a human body object that is substantially upright in the image. The human body of sport video is in a variety of ways during the movement, and the current detection algorithm is difficult to achieve the desired detection effect. In order to detect various postures from sport video of human targets. The paper proposed a human target detection algorithm without attitude limitation. Firstly, the algorithm finds the human candidate target by shape context matching. Secondly, the hypothetical target is removed by combining the direction gradient histogram features with the support vector machine. The experimental results show that the proposed algorithm not only can detect human targets in various poses, but also has a much higher detection speed than the direction gradient histogram detection algorithm.