Recognition of Dynamic Hand Gesture using Hidden Markov Model

A simple methodology for Malaysian Sign Language recognition using image processing is developed. Frame difference, thresholding of the frame difference and edge detection are used for the preprocessing stage. For the segmentation part, HSV color detection is used to detect the skin color. Combination of threshold of the frame difference, edge detection and HSV color detection are used to segment out the hand region. Feature extraction is used to track the hand gesture path. Centroid of the hand motion and 16-directional codewords are used. For classification, Hidden Markov Models (HMM) is used to recognize the hand gesture. A total of 100 videos are used for training and testing purpose. Overall, there are 14 gestures of the Malaysia Sign Language that were used in the recognition. The percentage of recognition for the testing set is 92.86%.