XR Picture Rendering With Multi-Moment Head Pose Prediction
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
XR devices face inaccuracies in predicting user head poses at future moments, leading to dizziness due to inaccurate rendering of images, as they rely on assuming constant inertial measurement data, which is not accurate when users are stationary at the current moment but moving at the future moment.
Innovation Solution
A method and apparatus that determine a series of predicted moments, including intermediate and display moments, input these into a pre-trained prediction model to obtain corresponding inertial measurement data, and use current pose and speed to accurately determine a display pose for rendering a virtual scene, thereby improving pose prediction accuracy.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If advance rendering is performed for XR devices, then images can be displayed normally at the corresponding moment, but the predicted pose at the future moment becomes inaccurate causing user dizziness
Solution Approach 1:
The patent performs preliminary rendering of images at predicted future moments by predicting user head poses in advance. The system determines a future moment earlier than the display moment, predicts the head pose at that future moment, and renders the image in advance based on the predicted pose, thereby ensuring normal display while addressing the accuracy issue through multiple predicted moments
Solution Approach 2:
The patent dynamically adjusts the prediction process by determining multiple predicted moments (first predicted moment, second predicted moment, etc.) between the current moment and the display moment. Each predicted moment has its own predicted pose and predicted image, allowing the system to adapt to changing user movements and select the most accurate prediction for display
2Device complexity
If inertial measurement data is assumed constant for prediction, then calculation is simplified, but accuracy deteriorates when users are stationary currently but moving at future moments
Solution Approach 1:
The patent segments the prediction process into multiple discrete predicted moments (first predicted moment, second predicted moment, third predicted moment, etc.). Each moment is processed independently with its own predicted pose and predicted image, allowing the system to handle different movement states at different time points without requiring complex continuous calculations
Solution Approach 2:
The patent changes the prediction parameters by using multiple different predicted moments instead of a single fixed prediction. Each predicted moment has its own inertial measurement data and pose parameters, allowing the system to adapt to changing movement conditions while maintaining manageable calculation complexity through the use of discrete time points
Data Source
AI summary
The present disclosure provides a picture rendering method and apparatus, the method comprising: determining a plurality of predicted moments after a current moment, wherein the plurality of predicted moments comprise a plurality of intermediate moments and a display moment after the plurality of intermediate moments; inputting the plurality of predicted moments into a pre-trained prediction model for prediction processing to obtain a plurality of predicted inertial measurement data that correspond to the predicted moments on a one-to-one basis; obtaining a current pose and a current speed of the head-mounted display device, and determining a display pose at the display moment based on the current pose, the current speed and the plurality of predicted inertial measurement data; and rendering a virtual scene based on the display pose to obtain a display picture corresponding to the display moment.