Abstract: Gestures recognition is of great importance to intelligent human-computer interaction technology, but it is also very difficult to deal with, especially when the environment is quite complex. In this paper, the recognition algorithm of dynamic and combined gestures, which based on multi-feature fusion, is proposed. Firstly, in image segmentation stage, the algorithm extracts interested region of gestures in color and depth map by combining with the depth information. Then, to establish support vector machine (SVM) model for static hand gestures recognition, the algorithm fuses weighted Hu invariant moments of depth map into the Histogram of oriented gradients (HOG) of the color image. Finally, an hidden Markov model (HMM) toolbox supporting multi-dimensional continuous data input is adopted to do the training and recognition. Experimental results show that the proposed algorithm can not only overcome the influence of skin object, multi-object moving and hand gestures interference in the background, but also real-time and practical in Human-Computer interaction.

Key words: gesture recognition, a weighted Hu, HOG , SVM, HMM