An action recognition method based on view-insensitive feature representation

In this paper, we propose a novel approach to represent and recognize human actions using the view insensitive representation in order to solve the problem of variations in viewpoint. This method could not only effectively combine the bag of interest point words in shot length-based video with the amplitude histogram of optical flow, but also solve the problem of the traditional bag of interest point words could not combine with Probabilistic Graphical Model which has a good capability of modeling the dynamic process of human action. The changed significantly frames found by the interest point detection strengthen the time relationship between frames, also greatly reduce redundant in the video so as to increase the recognition efficiency. At the same time this algorithm contains rich human motion information and anti-noise-interference ability because of the grid based radial bins histogram of optical flow, so that the accuracy rate was enhanced. At last HMM are used for modeling and recognizing human actions. The experiments on the IXMAS multi-view action dataset show that our approach has a satisfactory recognition result for variations in viewpoint.