Comparative analysis between two view-based methods: MHI and DMHI

In this paper, we compare the basic motion history image (MHI) and our developed multi-directional motion history image (DMHI) for human gesture recognition. One of the constraints of the MHI is that it erases past motion by overwriting new motion onto the past one, thereby creating a template that does not correspond the motion properly. We have solved this overwrite problem by employing the concept of motion descriptors from optical flow vector. We have separated the optical flow vector into four components based on the four directions, namely up, down, left and right. We have employed Hu moments to calculate the feature vectors for both the MHI and the DMHI methods. We have experimentally verified the superiority of the DMHI method in terms of recognition rate for complex motion. In this paper, we have also analyzed the importance of motion energy image for both methods, and with different motions, we have found that presence of energy image is more evident in the DMHI technique than in the MHI technique.