Real-time Driving Method And System Of Virtual Avatars Based On Kinect

As a new generation of Man-machine interaction platform,virtual reality focuses on immersive experience and emphasizes the depth rather than breadth of user connection and interaction.Avatar is the appearance that users use in a virtual scene.The application of 3D virtual avatars,which can be directly controlled by the user's body movements in real time,has always been considered as an important means to achieve deep immersion in virtual reality.The common solution to directly control the virtual avatar by the user's body movement is to capture the human motion in real time by the motion capture device,and then convert it into the virtual avatar movement,so as to realize the movement control of the virtual avatar.Nowadays,with the popularization of virtual reality environment,popular and low-cost virtual avatar driving method has been widely demanded,and the choice of low-cost and easy-to-use real-time motion capture equipment has become the main trend.In this study,Kinect was selected as the low-cost motion capture device to realize the real-time driving method of the virtual avatar,including real-time denoising of Kinect motion data and realization of the real-time driving system of the virtual avatar.The main research work of this study is as follows:1.A Kinect bone motion data denoising network was proposed,which consisted of a bidirectional recurrent autoencoder and a convolutional autoencoder in series.The bidirectional recurrent autoencoder was used to make the optimized data output by the network have higher position accuracy,while the convolutional autoencoder was used to make the optimized motion has more smoothly.The experimental results show that the proposed method is better than the existing denoising methods in position error,bone length error and smoothness error.2.On the basis of the denoising algorithm,the virtual avatar real-time driving system is proposed based on Unity and Python.The system can obtain the human skeleton motion data in real time through Kinect and with real-time denoising to drive the virtual avatar.3.Based on the virtual avatar real-time driving system,a motion somatosensory fitness game is implemented,and it is demonstrated that the proposed virtual avatar real-time driving method can effectively improve the recognition accuracy and anti-interference performance of motion sensing games.