Power-efficient method for estimating the user's gaze and eye features and a head-mounted eye-tracking device using the method
The method optimizes head-mounted eye-tracking devices by integrating a neural network-optimized edge computation unit on the nose bridge, reducing power consumption and latency through quantization, pruning, and architecture search, ensuring efficient and accurate gaze estimation.
Patent Information
- Authority / Receiving Office
- WO · WO
- Patent Type
- Applications
- Current Assignee / Owner
- VIEWPOINTSYST GMBH
- Filing Date
- 2025-08-28
- Publication Date
- 2026-05-28
AI Technical Summary
Existing head-mounted eye-tracking devices face high computational and power consumption issues, particularly in the pupil and feature detection stages, leading to increased latency, heat generation, and discomfort for the user.
A power-efficient method utilizing an edge computation unit with a machine learning hardware accelerator, employing quantization, post-training pruning, knowledge distillation, and neural architecture search to optimize neural networks for reduced computational resources and power consumption, integrated into a head-mounted device with eye sensors positioned on the U-shaped nose bridge.
The method achieves real-time user eye feature estimation with reduced power consumption, minimal latency, and lightweight design, maintaining accuracy while minimizing heat and interference with the user's nose and field of view.
Smart Images
Figure EP2025074578_28052026_PF_FP_ABST