Research On Static Hand Gesture Recognition Method Based On Depth Map

In recent years, human-computer interaction applications attracts more and more people's attention, which include Somatosensory Games, Sign Language Recognition System, Virtual Reality Experience and so on. Gesture recognition, which is one of the most outstanding technologies, has gradually become a very popular topic in the field of human-computer interaction. Earlier gesture recognition system was still based on traditional capture device, such as data gloves, which is not very convenient. Later, using color camera device to capture gesture becomes more popular. This method has high requirement in the light and the background, which limits its development. Until the appearance of the Kinect depth camera, it doesn't open a new chapter of the gesture recognition.The appearance of the Kinect opens a new chapter in the field of computer vision,which makes the gesture detection and segmentation much easier in the gesture recognition process. Researchers can concentrate more on the presentation of the gesture shape to improve accuracy. This technology has a very bright future. Making full use of the advantages brought by the Kinect can promote gesture recognition to a new level.Based on the existing framework of gesture recognition technology, the method of gesture feature selection algorithm, and the method of new gesture three-dimension projection and gesture area feature weighting fusion, which all have achieved good effects, are proposed respectively. Here are some specific research contents of this paper:1.An adaptive feature selection algorithm is proposed. Which feature descriptor to represent gestures by is an important research problem. Fixed feature descriptor is chosen in the recognition process of the current method. The author found that the distinguishing ability of different feature descriptors are related to their complexity. The feature descriptor, which has strong distinguishing capability, usually has higher complexity.Meanwhile, the difficulty of gesture recognition also has difference. Difficult gestures require feature descriptor that has strong distinguishing ability. And simple gesture canget better recognition accuracy by using simple feature descriptor. This paper propose that firstly we analyze the properties of gesture image to be recognized, then determine difficulty of the image, and represent the image by different feature descriptor. Thus, not only can the method ensure overall accuracy, but also reduce the time complexity and speed up the recognition process.2.New methods for gesture feature description are proposed. Firstly, a new gesture three-dimension projection technology is proposed. The cumulated value of the number of pixels with the same depth value is used as the new “GRAYSCALE” pixel value to represent projection result, which can retain richer gesture information. From three projection plane, feature descriptors are extracted respectively. Then in feature fusion process, a new method of feature weighting fusion is proposed, which gives a higher weight to a plane that has rich gesture information. Lastly, the experimental results show that the new method can improve accuracy for classification.