Real-Time Capable System for Hand Gesture Recognition Using Hidden Markov Models in Stereo Color Image Sequences

This paper proposes a system to recognize the alphabets and numbers in real time from color image sequences by the motion trajectory of a single hand using Hidden Markov Models (HMM). Our system is based on three main stages; automatic segmentation and preprocessing of the hand regions, feature extraction and classification. In automatic segmentation and preprocessing stage, YCbCr color space and depth information are used to detect hands and face in connection with morphological operation where Gaussian Mixture Model (GMM) is used for computing the skin probability. After the hand is detected and the centroid point of the hand region is determined, the tracking will take place in the further steps to determine the hand motion trajectory by using a search area around the hand region. In the feature extraction stage, the orientation is determined between two consecutive points from hand motion trajectory and then it is quantized to give a discrete vector that is used as input to HMM. The final stage so-called classification, Baum-Welch algorithm (BW) is used to do a full train for HMM parameters. The gesture of alphabets and numbers is recognized by using Left-Right Banded model (LRB) in conjunction with Forward algorithm. In our experiment, 720 trained gestures are used for training and also 360 tested gestures for testing. Our system recognizes the alphabets from A to Z and numbers from 0 to 9 and achieves an average recognition rate of 94.72%.