| The commercialization of VR technology in China was in 2015,and it was regarded as the “first year” of virtual reality(VR)in China.In recent years,the VR industry has been ups and downs.It is even a very novel concept for the ordinary consumer.And the popularity of VR in the main market has always faced great challenges.With the arrival of 5G commerce,the advantages of VR will be fully utilized,and its application in the VR industry is generally optimistic.The potential application development is also coming out.Unmarked 3D hand pose estimation is a fundamental challenge for many interesting applications of virtual reality(VR)and augmented reality(AR),such as object processing,game and interactive control in a VR environment.This task has been extensively researched over the past few years and has made great progress.Although many unmarked algorithms have achieved high accuracy under challenging conditions,most of the human body key point identification research starts from the general problem and is not considered under the VR application condition.Therefore,many excellent methods with high accuracy are not limited by the speed but limited by the accuracy,and the performance is mixed.So this paper studies the recognition and interaction of key points of 3D human body under the premise of VR application.The main work of the paper is summarized as follows:1.According to the gesture recognition algorithm,a dynamic gesture interaction method suitable for desktop VR is designed.Through the interactive experiment analysis of the meteorological data visualization system based on sitting posture VR,the designed dynamic gesture interaction method has a good effect in immersive meteorological data visualization system,providing users with a high degree of immersion and helping users to understand the data more effectively.2.For the shortcomings of Leap Motion gesture tracking application in desktop VR,we use computer vision technology combined with deep learning gesture recognition method to propose a strategy which calculates the self-occlusion degree of hand image captured by the depth camera.Simple gestures with low occlusion are identified by a fast model-based method.Complex gesture images use the trained Convolutional Neural Networks for key joint recognition,which not only ensures the recognition accuracy,but also improves the problem that the deep learning method is applied to VR with low number of frames.3.For the room VR application using VR handle,a low-cost fast human motion capture method is proposed.We decompose the human structure into the upper limb joint kinematic chain and the lower limb joint kinematic chain,and obtain the position information of the end joint from the head-mounted display and the handle trackers.The upper body motion is solved by inverse kinematics.The lower limb posture tracking uses the deep learning method to estimate the 3D leg's joint position from the color image captured by the camera,which improves the missing or inaccurate lower limb position information caused by the traditional inverse kinematic solution.Experiments show that our method only needs a color camera to reconstruct the whole body 3D posture in the room VR.Experiments show that the methods proposed in this paper for various specific VR environments improves the shortcomings of traditional general methods,and is more satisfied with the tracking requirements and operational experience of immersive virtual reality.Meanwhile,further improving the efficiency and accuracy of the recognition for human's key points under the virtual reality environment requires more consideration from mining human body's intrinsic structure feature. |
PDF Full Text Request