MOUNT: Learning 6DoF Motion Prediction Based on Uncertainty Estimation for Delayed AR Rendering.

Area of Science:

• Computer Vision
• Human-Computer Interaction
• Machine Learning

Background:

• Augmented reality (AR) rendering delays necessitate accurate head motion prediction to prevent virtual-physical world misalignment.
• Misalignment causes user discomfort, including time latency and dizziness.

Purpose of the Study:

• To propose a novel deep learning method for 6DoF (six degrees of freedom) motion prediction to compensate for time latency in AR.
• To improve the accuracy and smoothness of head motion prediction for enhanced AR experiences.

Main Methods:

• Developed a MOtion UNcerTainty encode decode network (MOUNT) utilizing deep learning.
• MOUNT estimates input data uncertainty and predicts output motion uncertainty.

Main Results:

• The proposed deep learning method significantly outperforms traditional hand-crafted approaches.
• Experiments demonstrated improved prediction accuracy and smoothness in head motion tracking.

Conclusions:

• The MOUNT network effectively addresses time latency issues in AR by enhancing motion prediction.
• The method significantly improves AR visual effects and user experience.