Eye Tracking Scan Region Expansion for Illuminance Instability
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Camera-based eye tracking technology experiences instability due to changes in illuminance and user motion, leading to failure in tracking eyes, particularly in driving environments with varying light conditions, necessitating a method to quickly reacquire eye positions and coordinates.
Innovation Solution
An image processing method that involves tracking a face region by using previous detected face regions, expanding the scan window size based on failed tracking attempts, and resetting the scan region to improve detection speed and accuracy, utilizing pupil center coordinates and feature points for alignment and tracking.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If camera-based eye tracking is used in driving environments, then viewpoint tracking and AR 3D HUD functionality are enabled, but tracking stability decreases due to illuminance changes and user motion
Solution Approach 1:
The system performs preliminary actions by maintaining a history of previous scan regions and detection results. When tracking fails, it quickly recalculates the scan region based on previously accumulated eye position information and historical scan regions, enabling rapid recovery without starting detection from scratch.
Solution Approach 2:
The system implements feedback mechanisms by continuously monitoring tracking success/failure and adjusting the scan region calculation accordingly. When tracking fails, it uses the failure information to recalculate and expand the scan region based on historical data, creating a closed-loop adaptive system that learns from past performance.
2Measurement precision
If the scan region is expanded to improve eye detection accuracy after tracking failure, then detection robustness increases, but processing time increases
Solution Approach 1:
The system performs preliminary calculations by maintaining historical scan region data and eye position information in advance. When tracking fails, it immediately retrieves and utilizes this pre-computed historical information to quickly recalculate the expanded scan region, avoiding time-consuming full-frame searches.
Solution Approach 2:
The system applies local quality by expanding the scan region selectively around the last known eye position rather than searching the entire image frame. This localized expansion based on historical data achieves robust detection while minimizing the search area and processing time.
3Reliability
If the scan window size is sequentially expanded based on failed tracking attempts, then detection coverage improves, but computational complexity increases
Solution Approach 1:
The system implements dynamics by making the scan window size adaptive rather than fixed. The window size is dynamically adjusted based on the number of consecutive tracking failures, expanding progressively (first window, second window, third window) until detection succeeds or a maximum is reached, allowing the system to respond flexibly to different failure scenarios.
Solution Approach 2:
The system changes parameters by modifying the scan region size and position based on historical eye position data and failure count. Instead of using a fixed scan region, it dynamically adjusts the region parameters (size, position, expansion direction) according to accumulated experience, optimizing detection coverage while managing computational load.
Data Source
Figure 1~2
Figure 3
Figure 4
AI summary
An image processing method includes acquiring an image frame; tracking a face region of a user based on first prior information obtained from at least one previous frame of the image frame; based on a determination that tracking of the face region based on the first prior information has failed, setting a scan region in the image frame based on second prior information obtained from the at least one previous frame; and detecting the face region in the image frame based on the scan region.