Viewpoint-oriented human activity recognition in a cluttered outdoor environment

Motion analysis in a cluttered outdoor scenario is a real challenging task. In this paper, a suitable human activity recognition approach is proposed that adapts with the noisy environment and with variable viewpoints. The recognition approach encompasses some motion processing methodologies having high precision for specific analysis, among those are: motion segmentation, optical flow computation, Motion History Image (MHI) generation, structured database development, and so on. Estimating the subject's moving body region, 4-directional MHIs of each activity are generated. An efficient pre-developed motion database aids in recognition of the activities. The proposed approach proves its significant precision in recognition for various activities from varying viewpoints.