Automatic and robust hand gesture recognition by SDD features based model matching

Automatic and robust hand gesture recognition remains challenging after many decades of study. Human beings are able to recognize a variety of hand gestures with 100% accuracy solely based on the contour of the hand. Hence, there must be an automatic method that is able to recognize the same variety of hand gestures solely based on the contour of the hand with 100% accuracy. The key technique lies in how to extract the features of the hand’s contour effectively. In this paper, we propose to recognize the hand gestures with the contour features extracted by slope difference distribution (SDD). Firstly, the hand is segmented, its centroid is computed and its contour is extracted. Secondly, the peak features and valley features of the hand contour are computed by the SDD. Thirdly, the hand gesture is recognized by model matching based on the SDD peak features and the SDD valley features. The proposed hand gesture recognition method was tested on three public datasets and it achieved 100% recognition accuracy for all the 10 gestures in two public datasets.