A Comprehensive Evaluation of Moving Static Gesture Recognition with Convolutional Networks

Static gestures is of great value for intelligent interaction. Most current researches on this field do not consider the complexity and randomicity of its practical applications. In this paper, we make a comprehensive study on moving static gesture recognition which is closer to the reality with Convolutional Neural Networks (CNN). First of all, Leap Motion (LM) is used as an acquisition equipment. Then we investigate the performance of networks in different depth and architectures. We design entirely new experimental scenes namely static gesture in motion. Finally, our proposed method is tested on a self-built database containing 12 different gestures and compared to built-in algorithm of LM in different recognition areas. Experimental results show CNN is robust and effective for static gesture recognition in the complex environment.