A Gesture Recognition Method Integrating RGB and Depth Image Features

Abstract

The key to the success of gesture recognition often depends on the extraction and representation of the gesture features. Compared with the traditional gesture recognition methods, the dual-channel convolutional neural network that we constructed in this paper can extract the features of RGB and depth images of the same gesture, making the gesture feature mining and extraction more accurate, and gesture recognition more robust. Firstly, the features are learned from the two channels respectively by convolutional neural network. Then, those features are mapped to the fusion layer, which the fusion features are used for classifier training. Finally, the model recognition rate can reach 98.11%, through the collected gesture database verification