Research On Improved Convex Shape Decomposition Hand Gesture Recognition

Throughout the development of computer,it has become more and more intelligent, and interaction between human and computer has become more and more natural which leads to the development of human-computer interaction. One of the most natural interactive mode is somatosensory interaction, people use their body languages to interact with computer. Hand gesture is one of the most abundant and most frequently used way of body language. However, there are still some shortcomings in the current technology of gesture interaction:the need to collect a large number of training data, not achieving real-time requirements and not identifying the similar gestures and so on. These shortcomings hinder the development of this technology, so it is still a challenging subject.In this thesis, several perspectives about hand gesture recognition are investigated, which includes hand segmentation, hand shape decomposition, extraction of hand skeleton. The main contributions of this thesis are summarized as follows:1)proposing a hand detection and segmentation method based on depth information and skin color information. According to the corresponding depth information of the skin color pixels of the background image labeled by Bayesian Decision Theory, we can build depth value histogram and get the depth threshold value. At last we segment hand from the background image with the threshold method, which is quick and accurate; 2) propose a improved convex shape decomposition algorithm for hand shape decomposition and get the hand skeleton. Compared with the other hand features, skeleton information can keep the original information of hand gestures to the maximum extent. And experimental results show that this method can lower the time complexity and reduce the computational complexity; 3) using a skeleton graph matching algorithm based on path similarity to recognize hand gesture. This method does not rely on the topological structure of hand, so it is more accurate and can increase the accuracy of the recognition of similar hand gesture. The feasibility and validity of the method is proved by a large amount of experiments.