Abstract: In order to solve the pro blem of poor gesture segmentation extraction, low image recognition rate and difficult recognition, a multi-feature fusion method was studied for fast gesture recognition. Firstly, a skin color distribution model was established based on the YCbCr color space model, removing the most of non-skin color interference from the complex background, so as to realize the gesture segmentation. And then, taking the texture image of the gesture ROI (region of interest), the shape image and the significant visual image as self-coding network input, the different types of features were linearly merged according as 5-layer stacked sparse autoencoders network framework. Finally, a SVM (support vector machine) classifier based on RBF (radial basis function) kernel function was used to classify the characteristic data, so as to realize the gesture recognition for different types of gestures. Experimental results show that, compared with other gesture recognition methods, the recognition rate is higher and the extraction characteristics are more representative. The average recognition rate can reach 95.05%.