An Improved Hidden Markov Model based on Human Motion Detecting Approach via Multi-View Image Sequences

In this paper, we concentrate on the problem of human motion detecting from multi-view image sequences based on a modified hidden Markov model, and human motion detecting is of great importance in computer vision. Firstly, the structure of human skeleton model is given, which refers to a local coordinate system. In this model, the human bones follow the Parent-Child relationship. In particularly, the upper node of human skeleton model is named skeleton root, and it can connect with the spine root. Apart from the spine root, the remainder parts of the human body can be extended from the leg, including: 1) thigh, 2) shin, 3) foot and 4) toes. Secondly, we locate the object of human body from multi-view image sequences, and then propose a novel method to find human’s head and shoulder. Thirdly, as a complete human motion representation may be the set of all threedimension points on a specific actor, in this paper, we represent human motion as four-dimension points in real environment. Next, a modified Hidden Markov model is designed to represent a symbol sequence, and then human motions can be obtained from the outputs of the proposed Hidden Markov model. To demonstrate the performance of the proposed method, a series of experiments are conducted. Experimental results show that compared with other existing methods, the proposed approach can effectively enhance the accuracy of human motion detection.