Research And Application Of The Static Gesture Recognition Based On The Kinect Depth Image

With the constant progress of science and technology and the popularity of computer, the human-computer interaction way, tends to be diversified. From the traditional keyboard, graphical user interface (GUI), tablets of handwritten Chinese characters to the recent hot speech recognition, gesture of somatosensory peripherals. Undoubtedly, the human-computer interaction way trends to be more nature and convenience. Gestures with the characteristics of intuition and naturalness have become an important means of human-computer interaction. It got rid of the bondage of traditional keyboard, mouse and so on, more in line with the habit of human beings, therefore has very broad application prospects. In this paper, gesture recognition is also chosen as the research subject.The traditional gesture recognition is mainly for gesture recognition based on data glove and gesture recognition based on vision technology. In gesture recognition based on data glove, users need to wear identification equipment, limiting the natural human-computer interaction. Gesture recognition based on vision research contains many algorithms, but these methods in the gesture at the time of the partitioning are susceptible to light and background, camera characteristics factors, which led to the recognition rate is not high.Based on the research of predecessors’ gesture recognition, this thesis has made the exploration research on gesture recognition technology. The main research idea is using the Kinect depth and color sensors to get the depth and color image data, combining with the method of gesture recognition based on vision, to analysis and identification of gestures. This method can solve well the influence of the factors of lighting, background. This article mainly for static hand gestures recognition. The research idea is to use the depth image histogram to find suitable threshold separation of foreground and background scene, and then use the color of skin information detection segment the gesture area. This method can extract the ideal hand area, and then extract the gesture area hog feature descriptor as feature vector. Lastly, use K neighbor (KNN) classifier as trainer. K neighbor in category decisions, with only very small amounts of the adjacent samples, therefore, K neighbor can avoid the problem of unbalanced samples. Use gesture recognition algorithm proposed in this paper for five kinds of common static gestures to experiment, and implements a small gesture recognition system, at the end, this paper has carried on the experiment and the experimental data analysis, from the recognition rate of the algorithm and the change of light and rotation, translation, scale conditions on the result of recognition aspects such as the corresponding conclusion, thus verified the feasibility and robustness of the gesture recognition method.