Aiming at the shortcoming that a single gesture feature cannot describe the gesture well in hand gesture recognition, a hand gesture recognition method based on Hu moments and GrayLevel Cooccurrence Matrix (HuGLCM) is proposed. Firstly the skin color model is used to segment the captured image to get the gesture area. Secondly, the single connected contour of the gesture is extracted by mathematical morphology and polygon fitting. The improved HuGLCM method is used to extract the geometric shape and texture features of the gesture and establish a template database. Finally, the gesture image is identified and classified by the extended Canberra distance. The experimental results show that the improved algorithm has an average recognition rate of more than 95% on seven kinds of gestures, and the computing speed is fast, which can meet the realtime requirements.