Studies On The Gesture Recognition System Based On Computer Vision

The hand gesture recognition based on computer vision is a key technology to thenext generation of human-computer interaction. This thesis presents a vision-basedgesture recognition algorithm from points of hand segmentation, feature extraction andgesture recognition.In the part of hand segmentation, the thesis proposes a method which uses bothmotion and skin-color detection to separate the hand from cluttered backgrounds. First,we use motion detection to locate the moving region, and then we use skin-colordetection to separate the hand. After that, to get better results, we take measures toerase noises in the image of the hand gesture. Finally, we get the binary imageincluding hand gesture only.In the part of feature extraction, according to the characteristic of the hand, thethesis chooses the Hu invariant moments and circular degree as the feature vector ofthe hand gesture. The results show that the feature is robust to noise and has goodinvariability to translation, rotation and proportional transformation.In the part of gesture recognition, this thesis proposes a new recognition algorithmbased on Support Vector Machine (SVM) and the nearest neighbor algorithm. First, theSVM is trained with a large number of samples. After the SVM is trained successfully,it can be used to recognize the hand gesture. To the condition of wrongly classified thatone gesture belongs to several classes, the nearest neighbor algorithm is used in gesturerecognition to raise recognition rate.Based on the algorithm, this thesis provides a video-based gesture recognitionsystem. The test shows that this system has higher recognition rate and better real-time.