A Hand Gesture Recognition Algorithm And Its Implementation Based On Monocular Vision

Human-Computer Interaction (HCI) is an eternal subject in the computer field, how to build a harmonious and natural environment are the hotspot and difficulty in the current human-computer interaction. The hand gesture, as a natural and intuitive interaction, meets the people's communication habits, attracting many researchers' interest. Currently, the hand gesture recognition technology can be classified into two major categories in terms of different input devices:Data glove based and Vision based. Vision-based hand gesture recognition directly uses hands as the human-computer interaction manual input device so that the interaction seems more natural and easy, becoming the research focus of the current hand gesture recognition, which is the key technology to realize a new generation of human-computer interaction. The major drawbacks of the current vision-based gesture recognition are:the low recognition rate, the poor real time and the sensitivity on the illumination change, the scale transform and the plane rotation. Using single hand gesture based on monocular vision as the research object, the paper discussed a hand gesture recognition method that meets the recognition rate and the real-time requirement, and designed and implemented a gesture recognition system. The vision-based gesture recognition process can be divided into the hand tracking, the hand gesture segmentation, the feature extraction of hand gesture, the hand gesture modeling and the hand gesture recognition. Based on the existing technology, the paper completed the main work as follows.(1) According to the structure of the gesture recognition system, it introduced in detail the common techniques in each stage and by comparing and analyzing through a series of experiments, pointed out its merits and demerits. In practical, the appropriate algorithm should be selected in terms of the system requirement. The gesture recognition system requires ensuring a high recognition rate and meets the real-time requirement. Thus, the methods selected in the paper are all looking for the high recognition rate and real-time balance.(2) It proposed a fingertip detection method based on the convex defect. The traditional detection methods require retrieving and scanning the palm's outer contour, which causes a lot of calculation. The paper pointed out that it can effectively represent the gesture with the convex defect. Using the convex defect to detect the fingertips can effectively make up the current poor real-time detection of fingertips, and also ensure the accuracy of fingertips detection.(3) Based on the analysis and comparison of the existing hand gesture recognition technologies, the paper proposed a multi-features hand gesture recognition method with one single camera, according to two aspects:hand gesture segmentation and modeling. First, using the color features to detect the skin color region, through the CamShift algorithm it realized automatic tracking and rapid positioning of hands, and then in YCgCr color space managed to segment hands. Secondly, use the hand contour and the convex defect to detect fingertips, and the number and orientation of fingertips to represent a gesture, and then complete the hand gesture recognition with geometric features such as the contour length and the area. Experimental results show that the method proposed is robust and real-time, being able to adapt to the environmental changes.