Disclosed are a human gesture recognition system and method, belonging to the technical field of intelligent controlled and attended human gesture recognition. The system and the method solve the problems that the human body is not obviously different from the background characteristic and is difficult to recognize during image video recognition on human motion gestures and the target cannot be effectively and timely tracked in area-oriented recognition. According to the human gesture recognition system and method, acquired information is preprocessed to generate relatively pure and available data stream. Frame-by-frame feature extraction is performed on the pure and available data stream to obtain an input sample of subsequent training. A classifier is trained by using the extracted feature. Information is acquired by the same acquisition module and input to the trained classifier for classified recognition. The recognition accuracy reaches 98%, and the system and the method are suitable for all devices that acquire human motion information by adopting wrists and legs.