Robustness Research Of Gesture Recognition Technology Based On Structure Light

As the development of computer technology, the interaction between human and computer is becoming more and more simply and humanize. As a natural means of human-computer interaction, gesture causes researchers'attention widely. Gesture recognition technology mainly studies the detection and recognition of hand gestures, implements the anthropomorphic human-computer interaction, and can apply in robot control, sign language recognition, etc.At present, gesture recognition technology mainly includes the method based on wearable sensors and the method based on visual. The gesture recognition technology based on wearable sensors need to wear a senor device which is similar to the gloves. This approach sometimes can bring inconvenience to users (for example sweating), and the device is more expensive. However, gesture recognition technology based vision use camera to capture gestures non-contact. This is a more natural way of interaction. Traditional gesture recognition technology based vision captures the two-dimensional image of gestures, which is vulnerable to the influence of background, light, angle, etc. By improving traditional gesture recognition technology based vision, this paper proposes a gesture recognition technology based on structured light. Which uses a 3D depth camera capturing the depth data of gesture by scanning the gesture, further constructs the point cloud data of gesture surface, then filter 3D point clouds based depth data and extract gesture's region of interest (ROI) using the latest target method. And then the skin color detection technique was used for gesture positioning. In the end, the Adaboost algorithm was used for gesture recognition, future extract gesture area. In order to improve the robustness of the measurement method, this paper defines two kinds of hand shape which are fist and palm. The recognition rates of two gestures are both better. By these three steps that the extraction of region of interest, skin color detection and Adaboost identification, this paper accurately extracted the gesture region of fist and palm. Next, in order to realize the recognition of dynamic gestures, this paper defines the five gestures that up, down, left, right and check mark, then implies HMM mode for the recognition of gestures. The recognition rates of five gestures are all better.The test results show that 3D depth camera can quickly obtain the depth image data and not consider too much about light and texture environment; Through these three steps that the extraction of region of interest, skin color detection and Adaboost identification, this paper accurately extracted the gesture area of fist and palm, and improve the robustness of the system; The HMM mode is used to recognize gestures of up, down, left, right and check mark, and basically achieve the function of real time. These results shows that the paper's algorithm is fast, accurate, high resolution, good anti-interference, good robustness, easy to extract the gesture area, good real-time, and has the advantages of simple structure, low cost and easy to realize.