Adaptive Eye Tracking Model Engine for Vehicle Gaze Estimation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional eye tracking technologies are limited in surround or 3-D environments, such as vehicles, due to differences in head pose and gaze angles compared to non-surround environments, and struggle with varying illumination conditions and accuracy in pupil movement estimation.
Innovation Solution
An adaptive eye tracking machine learning model engine that collects and processes data from surround scenes using multiple sensors and perspectives to generate customized eye tracking models for specific deployment environments, including vehicles, using techniques like Deep Neural Networks and facial landmark neural networks.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If conventional SVM gaze classifier with spatial configuration is used, then the system can classify gaze regions with fixed vehicle geometry, but the accuracy varies between subjects and within subjects and does not obtain explicit pupil features
Solution Approach 1:
The patent transforms the gaze classification approach from spatial configuration-based parameters to pupil movement-based parameters. By tracking pupil position, diameter, and velocity, the system captures dynamic ocular features that are more sensitive to gaze direction and less variable across subjects, thereby improving measurement precision while maintaining adaptability.
Solution Approach 2:
The patent replaces the mechanical/spatial configuration-based classification system with a physiological-based system that measures actual pupil dynamics. This substitution allows the system to directly measure gaze-related physiological responses rather than inferring from spatial relationships, improving accuracy while accommodating subject variability.
2Measurement precision
If CNN trained with facial descriptors is used, then the model can classify gaze regions, but the validity and accuracy are limited to specific vehicle types with fixed 3-D geometry
Solution Approach 1:
The patent creates a universal eye tracking model that can operate across different vehicle types and geometries. By focusing on pupil movement patterns rather than vehicle-specific spatial configurations, the system achieves multi-functionality that adapts to various deployment environments including different vehicle architectures, making the solution universally applicable.
Solution Approach 2:
The patent extracts the essential gaze-tracking functionality from vehicle-specific constraints. By isolating and focusing on pupil movement as the primary measurement, the system removes dependency on fixed vehicle geometry and spatial configurations, enabling the model to be transferred across different vehicle types.
3Adaptability or versatility
If conventional eye tracking is used in surround environments, then the system can operate in 2-D environments, but it struggles with wider head pose and gaze angles and varying illumination conditions in 3-D surround environments
Solution Approach 1:
The patent implements dynamic adaptation to varying illumination conditions by continuously tracking pupil characteristics that are less sensitive to lighting changes. The system adjusts its measurements based on real-time pupil diameter and velocity data, maintaining reliability across different illumination scenarios from bright sunlight to darkness.
Data Source
AI summary
In various examples, an adaptive eye tracking machine learning model engine (“adaptive-model engine”) for an eye tracking system is described. The adaptive-model engine may include an eye tracking or gaze tracking development pipeline (“adaptive-model training pipeline”) that supports collecting data, training, optimizing, and deploying an adaptive eye tracking model that is a customized eye tracking model based on a set of features of an identified deployment environment. The adaptive-model engine supports ensembling the adaptive eye tracking model that may be trained on gaze vector estimation in surround environments and ensemble based on a plurality of eye tracking variant models and a plurality of facial landmark neural network metrics.