Gaze Point Estimation Using Neural Network and Optical Flow
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional eye tracking systems require multiple apparatuses and struggle to accurately estimate gaze points, especially in environments where subjects are moving or blended into crowds, leading to degraded estimation accuracy.
Innovation Solution
A gaze point estimation processing apparatus that uses a storage unit to store a gaze point estimation model generated through deep learning, allowing processors to estimate gaze points from images using this model, independent of external apparatuses like cameras, and considers attributes and optical flow for improved accuracy.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If multiple apparatuses are used for eye tracking, then gaze point estimation can be performed, but device complexity increases
Solution Approach 1:
The patent combines multiple eye tracking apparatuses (camera for photographing eyes, head orientation detection apparatus, distance measurement apparatus) into a single integrated eye tracking system. This merging allows the system to perform gaze point estimation using a single apparatus while maintaining the functional capabilities of multiple separate devices, thereby reducing device complexity without sacrificing estimation reliability
Solution Approach 2:
The integrated eye tracking apparatus performs multiple functions that would traditionally require separate devices: it photographs eyes to detect feature points, detects head orientation, measures distance to reference points, and estimates gaze points. This multi-functionality allows a single apparatus to replace multiple specialized devices, resolving the contradiction between reliability and device complexity
2Measurement precision
If conventional eye tracking is used, then gaze point can be calculated in controlled environments, but estimation accuracy degrades when subject is moving or blended into crowd
Solution Approach 1:
The patent segments the gaze point estimation process into multiple independent components: feature point detection from eye photographs, head orientation detection, distance measurement to reference points, and final gaze point calculation. This segmentation allows each component to be optimized independently and ensures that the overall system maintains accuracy even when some components face challenges in moving or crowded environments
Solution Approach 2:
The patent introduces multiple intermediary measurements and calculations: feature points of eyes serve as intermediaries to detect eye position and orientation; head orientation detection serves as an intermediary to compensate for head movement; distance measurement to reference points serves as an intermediary to calculate three-dimensional gaze position. These intermediaries enable accurate gaze point estimation even when subjects are moving or blended into crowds by providing additional reference information
Data Source
AI summary
A gaze point estimation processing apparatus in an embodiment includes a storage configured to store a neural network as a gaze point estimation model and one or more processors. The storage stores a gaze point estimation model generated through learning based on an image for learning and information relating to a first gaze point for the image for learning. The one or more processors estimate information relating to a second gaze point with respect to an image for estimation from the image for estimation using the gaze point estimation model.