Study On Techniques Of Hand Gesture Recognition Based On Monocular Vision

With the development of science and technology, human-computer interaction is playing an important role in the daily life. Because the intuitive and natural characteristics, hand gesture has become an important means of information exchange between the people and people. As the one of the most important human-computer interface technique, vision-based gesture recognition is a hot topic in pattern recognition.This thesis designs and implements both static and dynamic gesture recognition system based on the single-camera. The static gesture recognition system can capture and recognize15types of common static gestures and the dynamic gestures system can recognize10types of common dynamic numeric gesture in real-time.The main works of this thesis are as follows:(1) Introduces the current situation and basic principle of gesture recognition, such as gesture classification, class space and so on.(2) Presents a gesture detection method based on combination of frame difference and skin color. Then it uses CamShift tracking algorithm for gesture tracking to improve the efficiency of the system.(3) Presents a gesture segmentation method based on Gaussian Mixture Model in YCgCr color space. Then it uses dynamic threshold to binarize the likelihood image. Experiments show that the algorithm improves the accuracy of the gesture segmentation.(4) For the design of static gestures, the paper presents to extracts the joint features of Fourier descriptor, Hu moment invariant and hand gesture area feature. Then it selects Support Vector Machine (SVM) as classifier to recognize the gesture. It is shown that the static gesture recognition system in the paper has a high recognition rate.(5) The thesis uses the static gestures to mark the start and end of dynamic gesture. It proposes a rapid fingertip detection method based on combination of curvature and centroid distance. Then it extracts the change in the angle of the fingertip trajectory and selects Hidden Markov Model to achieve trajectory recognition. The simulation results show that the system renders fine recognition rates and guarantees strong robustness and instantaneity.