Hand Gesture Segmentation Algorithm Based On Vision

With the appearance of the Internet of things and the cyber physical systems (CPS),Human-computer interaction technology is more important in our lives. Gesture, as oneof the most natural and important communication method between human and other people,has already been a new way for human-computer interaction (HCI).Gesture recognition based on vision contain six stages, they are gesture image acquisition,gesture segmentation, gesture tracking, feature extraction and dynamic gesture recognition.The critical step about the hand gesture recognition is the gesture segmentation, it will have adirect impact on the system’s recognition rate. The main feature of the gesture segmentation isto rationally and effectively select the characteristics of the interested gesture region.Traditional segmentation algorithm couldn’t receive the good result in a complexenvironment. Therefore this paper propose several methods to solve the problem that theaccuracy and efficiency could not be positively related to each other in traditional handgesture segmentation. Firstly, one of method is a improved active contour model is proposedto break up the hand, we increase the energy of the shape constraint a nd the actual shape ofthe gesture based on the traditional active contour model. Improved active contour modelcould shrink to the depression area of the gesture in order to obtain the better result of thegesture segmentation because of expanding the scope of the contour points.Another new method is hand gesture segmentation based on improved kalman filter andTSL skin color model. This method use improved kalman filter to pre-judge the hand positionfor the sake of decreasing the interference of the similar skin color at first. Secondly, use TSLskin color model to realize hand gesture regions. Finally, holes and isolated small targets areremoved by morphological approach. This method could reduce the interference of the similarskin color regions compare with the traditional skin color segmentation.The experimental results indicate that our methods are capable of segmenting thegestures effectively and could be used in real-time video processing.