Optical Flow guided Motion Template for Hand Gesture Recognition

Abstract

Gesture representation specially hand gesture has a special role in the computer and human interaction community. Model-based and appearance-based methods are two primary techniques for hand gesture representation. Apart from these two, space-time features and motion-based approaches have gained quite impressive performance in various applications of action and gesture recognition. In space-time features, actions/gestures are considered as local spatiotemporal neighbourhood. But most space-time features are computationally expensive. Motion-based approaches mainly constitute optical flow and motion templates. Motion estimation of the image pixels is the key factor in optical flow, whereas, in motion-templates, video-wide temporal evaluation and their representations are widely used for action/gesture recognition. Both these methods have their own advantages and accordingly applied in the analysis of motion and related applications. In this paper, we tried to combine both and proposed a new method to get the advantages of individual methods in representing the temporal templates of a video by fusing the video dynamics into a single image. The main benefits of the technique are basically its simplicity, ease of implementation, competitive performance and efficiency.