Dynamic Gesture Trajectory Recognition Based On HMM

Vision-based hand gesture recognition is an important part of the new generation of human-computer interaction and the robot application,it has very important research value and research significance.In this thesis,the basic theory and the background of the vision based gesture recognition are summarized,five common dynamic gestures are used as the basic object of recognition.In order to simplify the background in the experiment,the main consideration is the human face,the human arm and the objects which are close to skin color in the background.According to the requirements of the subject,the whole gesture recognition system can be divided into three parts:the hand locating and tracking of the hand,feature extraction of the trajectory and the trajectory recognition.In the part of hand locating and tracking of the hand,the image is converted from RGB color space to YCbCr color space to extract the skin color region,an algorithm based on two consecutive frames subtraction and background subtraction is used to detecte the movement gesture,the particle filter algorithm is used to track the moving gesture.A method of using partial propagation instead of global propagation is proposed,which can include the moving object and reduce the number of particles in tracking to reduce the computational complexity.In the part of the feature extraction of the trajectory,we use the direction code as the feature of the trajectory,the moving angle is divided into 12 equal parts.According to the movement of the angle,we give a corresponding encoding,which is stored in a sequence.In the end,we can get a vector sequence.In the part of the trajectory recognition,the Baum-Welch algorithm is used to train five dynamic hand gestures,the forward probability of the trajectory vector sequence is calculated by forward algorithm,and the maximum probability of HMM is the result.In order to verify the above algorithm,a hand gesture recognition system is designed.This system can use hand and machine to perform simple trajectory recognition.Experiments show that the tracking efficiency of this system is high,and the five nredefined gestures can also be effectively identified.