Gaussian Gaze Zone Detection for Head Pose and Kappa Variation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing gaze estimation technologies face challenges in accurately determining gaze direction and zone due to issues with head pose estimation at large rotation angles, occlusion of facial landmarks, unstable tracking of arbitrary features, and neglecting eye-variation parameters like the kappa angle, leading to inaccurate gaze zone detection.
Innovation Solution
A Gaussian-based method for gaze zone detection that estimates gaze direction using a 3D face model, adjusts yaw and pitch angles for consistent head location, and applies Gaussian distributions to associate gaze directions with predefined zones, compensating for individual variations in kappa angles.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If traditional gaze estimation methods are used, then the system can operate with simple processing, but accuracy deteriorates due to head pose estimation failures at large rotation angles
Solution Approach 1:
The patent transforms the gaze estimation problem from a complex 3D head pose dependent problem into a simpler 2D image coordinate system problem. By representing gaze direction as angles in a camera coordinate system rather than relying on complex head pose estimation, the system achieves better accuracy without proportionally increasing processing complexity.
Solution Approach 2:
The patent introduces an intermediary coordinate transformation approach. Instead of directly estimating gaze from complex head poses, it uses an intermediate representation (gaze angles in camera coordinate system) that bridges the gap between simple image processing and accurate gaze estimation, avoiding the need for complex head pose algorithms.
2Reliability
If facial landmarks are used for gaze estimation, then the method can be implemented with standard image processing, but reliability deteriorates due to occlusion of landmarks
Solution Approach 1:
The patent creates a simplified copy of the gaze estimation problem in a different coordinate system. Instead of tracking facial landmarks directly (which are subject to occlusion), it uses an intermediate representation of gaze direction as angles in the camera coordinate system, which can be derived from eye position and head orientation without requiring direct landmark tracking.
Solution Approach 2:
The patent segments the gaze estimation into independent components: eye position detection, head orientation estimation, and coordinate transformation. This segmentation allows each component to be optimized independently and reduces the impact of occlusion on the overall system reliability.
3Adaptability or versatility
If arbitrary features are tracked for gaze estimation, then the system can handle various head positions, but measurement precision deteriorates due to unstable tracking
Solution Approach 1:
The patent changes the parameters used for gaze representation from arbitrary feature-based coordinates to standardized angular parameters (yaw and pitch angles) in the camera coordinate system. This parameter transformation provides both adaptability to different head positions and stable, precise measurements because the angular representation is invariant to the specific coordinate system used for feature tracking.
4Manufacturing precision
If kappa angle variations are ignored, then the estimation method remains simple, but manufacturing precision deteriorates due to individual eye variations
Solution Approach 1:
The patent addresses kappa angle variations by transforming the gaze estimation into a coordinate system where these anatomical variations are naturally compensated. By using camera coordinate system angles and Gaussian distributions to model gaze zone associations, the system accounts for individual eye variations without requiring complex subject-specific calibration or models.
Data Source
AI summary
Certain aspects of the present disclosure provide techniques for gaze zone detection. A method generally includes estimating a first gaze direction of a user based on at least one or more first two-dimensional (2D) images of a face of the user, wherein the first gaze direction is represented as a first yaw angle and a first pitch angle in a camera coordinate system; and associating the first gaze direction of the user with a first gaze zone among a plurality of gaze zones based on: the first yaw angle; the first pitch angle; and a first Gaussian distribution of the first gaze zone, wherein the first Gaussian distribution is based on a plurality of gaze directions, of a plurality of users, associated with the first gaze zone.