Gesture Recognition and Localization Using Convolutional Neural Network

Gesture recognition based on computer vision, which is natural and intuitive, is getting more and more attention in the field of human-computer interaction. Due to the limitations of existing gesture recognition methods, this paper presents an efficient and effective solution, which is divided into two main parts: dataset collection and the convolutional neural network (CNN) design. Firstly, a novel method is proposed to collect the hand gesture dataset through the sliding window, which makes it easy to capture arbitrary gesture data in any background, and the whole process is simple and fast. Secondly, a novel CNN named HandNet is designed for improving the performance of the gesture recognition and localization. The bone structure of the proposed CNN is HandBlock, which has the ability of richer feature expression and adaptive weight expression between high-level features and low-level features. The experimental results show that the proposed model achieves comparable performance with the state-of-the-art methods.