Studieson The Dynamic Gesture Track Recognition System Based On Monocular Vision

In recent years, with the development of computer vision technology, the dynamicgesture recognition based on vision has become one of the important researchdirections of the multiple pattern man-machine interface technology and received wideattention of home and abroad. This thesis presents a dynamic gesture track recognitionalgorithm based on monocular vision from points of gesture detection andsegmentation, dynamic gesture tracking and dynamic gesture track recognition.In the part of gesture detection and segmentation, the thesis use a method, whichis the combination of motion detection, skin detection and Gabor texture detection.First, we use the motion detection to locate the candidate gesture region, and then weuse skin detection in the candidate gesture region to get the area of candidate skin.After that, we use Gabor texture detection in the candidate skin region. Finally, to getthe better results of gesture detection and segmentation, we take morphologyprocessing with the segmentation results.In the part of dynamic gesture tracking, the algorithm, which is the combinationof Kalman filter and Mean Shift tracking algorithm, is used in this paper. Using theKalman filter, this algorithm can forecast the position of the dynamic gesture and thewidth of the tracking window. To achieve the purpose of real-time tracking, we put thepredicting outcomes, as an input, into the Mean Shift tracking algorithm to track thedynamic gesture.In the part of dynamic gesture recognition, this paper makes a study of HiddenMarkov Model (HMM), and realizes the training algorithm of gestures model and therecognition algorithm of gesture track, which based on the Hidden Markov Model.Finally, with the algorithm research above, a dynamic gesture recognition systembased on the video streaming has been designed and realized in this thesis. Theexperimental results shows that the system has a higher recognition rate.