A Static Gesture Recognition Method Based on Improved SURF Algorithm and Bayesian Regularization BP Neural Network

Gesture recognition plays an important role in the aspect of human computer interaction (HCI). It has become one of the most challenging tasks in the pattern recognition field. So far, many gesture representations using two-dimensional image have been proposed, but normally they are vulnerable to environmental factors, such as illumination, cluttered backgrounds and so on. In this paper, we propose a static gesture recognition method based on the improved speed up robust feature (SURF) algorithm and the Bayesian regularization back propagation (BP) neural network with the Microsoft Kinect sensor. With the advantages of the Kinect, we can capture the depth data to enhance the robustness of the proposed algorithm. Gesture analysis can be viewed as a two-fold problem, i.e., gesture representation and classification. On the one hand, we implement gesture segmentation by the depth data, and then extract the feature descriptor of the gesture based on the improved SURF algorithm which is optimized through the key point detection and orientation calculation. On the other hand, the method based on the Bayesian regularization BP neural network is used as classifier. Subsequently, in order to further intensify the recognition accuracy, another method of classification of gestures based on maximum angle between fingers is proposed as well. Finally, two kinds of classification results are also combined to get the final classification result. The experimental results show that the proposed method can eliminate the interference of the background, and enhance the robustness and accuracy of the gesture recognition.