查看文章

In this paper, we propose a histogram based feature to represent and recognize human action in video sequences. Motion History Image (MHI) merges a video sequence into a single image. However, in this method, we use Directional Motion History Image (DMHI) to create four directional spatiotemporal templates. We, then, extract the Local Binary Pattern (LBP) from those templates. Then, spatiotemporal LBP histograms are formed to represent the distribution of those patterns which makes the feature vector. We also use shape feature taken from three selective snippets and concatenate them with the LBP histograms. We measure the performance of the proposed representation method along with some variants of it by experimenting on the Weizmann action dataset. Higher recognition rates found in the experiment suggest that, compared to complex representation, the proposed simple and compact representation can achieve robust recognition of human activity for practical use.