Research On Continuous Dynamic Hand Gesture Recognition

Human Computer Interaction (HCI) plays a very important role in all kinds offields. Acting as the best visualized, imagery, vivid means of the HCI, consequentlyhand gesture becomes one of the hot topics in the HCI research. This paper uses a singlecamera to recognize the single hand movements. Therefore, this paper entitled researchon continuous dynamic hand gesture recognition mainly to solve following problems.The problems are that the light, face, arm and color similar to skin color interfere withthe palm segmentation, and for continuous hand gesture recognition how to separate thepredefinedisolate hand gesture with the non-meaningful movements and how todetermine the start and the end of a hand gesture. In order to complete this subject andsolve the problems mentioned above, this paper has conducted the research on thecorresponding algorithm and carried out a number of experiments. The experimentsshow that the method to solve the corresponding problem proposed in this paper iseffective and achieves good results. The hand gesture recognition process includes threekey links which are hand gesture segmentation, feature extraction and hand gesturerecognition. So the research work and proposed method in this paper are followed:1) Fusion of skin color detection and motion detection as the palm segmentationmethod is proposed to effectively solve the problem that the light, face, arm and colorsimilar to skin color interfere with the palm segmentation in the complex background.In this paper, image preprocessing is performed first to make the light balance anddecrease the noise. Then it combines the YCbCr skin color detection with the singleGaussian background difference method to remove the relatively static face and colorsimilar to skin color and further get the rough hand area. Finally it uses the post-processing method of opening and closing operation in morphology and connectedregion labeling to remove the redundant small regions and fill the holes and even get thecomplete hand area including arm.2) Direction chain code as the feature extraction method of hand gesture trajectoryis presented to greatly reduce the amount of calculation. After the palm segmentation, itgets a series of centroid points of palm regions. Then by connecting the centroid points,a hand gesture motion trajectory is obtained. According to the coordinates of twoadjacent centroid points, it can calculate the angle of the two centroid points. This paperuses12direction chain code to code these angle with these numbers from1to12andfinally get the one-dimensional eigenvector.3) A threshold model based on Hidden Markov Model as the continuous dynamichand gesture recognition method is proposed to separate the predefined hand gestureand non-meaningful movements, as well as to determine the beginning and the end of ahand gesture. This paper firstly recognizes the predefined isolated dynamic handgestures by HMM to get the robust HMMs of each predefined isolated dynamic handgesture. Then by merging the robust HMMs of each predefined isolated dynamicgesture into a model, the threshold model is obtained. Finally it uses the thresholdmodel and the robust HMMs of each predefined isolated dynamic gesture to recognizethe continuous dynamic hand gesture.