Two-Person Interaction Action Recognition Based on Multi-Source Information Fusion Algorithm

The existing methods of two-person interaction action recognition based on RGB image is greatly affected by illumination change, object occlusion and environmental change. Considering the respective advantages of the RGB image and the depth image, and the characteristics of information complementarity, this paper proposed a multi - source information fusion algorithm. In our proposed method, the recognition probability of the RGB image and the depth image are weighted fused for the two-person interaction action recognition. Firstly, the frame difference method and ViBe algorithm are respectively used for moving object detection and segmentation. Secondly, histogram of oriented gradient (HOG) features are respectively extracted from the moving regions of the RGB image and the depth image. Thirdly, the nearest neighbor classifier algorithm is used to recognize the actions of the RGB image and the depth image. Finally, the recognition results of the RGB image and the depth image are weighted fused. Experimental results show that the method achieves the better recognition rate.