An Approach of Delay Data Processing in HMD Based Training Systems

摘要

This paper presents a method of processing the delay data in virtual reality training systems based on head-mounted display (HMD). The delay data is a time series and can be predicted. We propose to employ neural network as a tool to predict the next head position and orientation that can help computing and updating the display faster, and reducing or eliminating dizziness caused by delays. The network input is the value of variable to be predicted at one or more previous time steps. The output is the prediction of its value at the next time step. Flock of Birds was employed as the tracking system of our HMD based training system. Eight types movements of the head were defined to provide representative training and testing sets for the learning system. We chose head positions x, y, z and head angles α , β, γ as the inputs of the neural network, which provide a good representation of the dynamics and characteristics of the raw data sets. The network utilized is a Recurrent Neural Network (RNN) which is a type of Discrete-Time Recurrent Multi-layer Perceptions. The RNN training algorithm we chose supports off-line supervised learning. The results demonstrate that the delay data occurred in our HMD based virtual reality training system can be processed effectively.