This paper proposes a hand interface through a novel deep learning that provides easy and realistic interactions with hands in immersive virtual reality. The proposed interface is designed to provide a real-to-virtual direct hand interface using a controller to map a real hand gesture to a virtual hand in an easy and simple structure. In addition, a gesture-to-action interface that expresses the process of gesture to action in real-time without the necessity of a graphical user interface (GUI) used in existing interactive applications is proposed. This interface uses the method of applying image classification training process of capturing a 3D virtual hand gesture model as a 2D image using a deep learning model, convolutional neural network (CNN). The key objective of this process is to provide users with intuitive and realistic interactions that feature convenient operation in immersive virtual reality. To achieve this, an application that can compare and analyze the proposed interface and the existing GUI was developed. Next, a survey experiment was conducted to statistically analyze and evaluate the positive effects on the sense of presence through user satisfaction with the interface experience.
DeepHandsVR: Hand Interface Using Deep Learning in Immersive Virtual Reality
