Gaze Point Determination via 3D Eye Model Segmentation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current face recognition technologies rely heavily on geometric information and deep learning for precise positioning of face feature points, but they do not effectively determine gaze points, which are crucial for understanding eye state and behavior analysis.
Innovation Solution
A method and apparatus for determining gaze points by obtaining two-dimensional coordinates of eye feature points, including an eyeball center area feature point, and converting them into three-dimensional coordinates within a preset coordinate system to accurately calculate the gaze point position using neural networks and machine learning techniques.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If face recognition uses deep learning for positioning face feature points, then positioning precision is improved, but the system cannot effectively determine gaze points
Solution Approach 1:
The patent segments the face recognition task into two distinct parts: (1) face feature points positioning using deep learning, and (2) gaze point determination using a separate geometric model. By dividing the system into these independent modules, each can be optimized for its specific function without interfering with the other, thus preserving gaze point information while maintaining positioning precision.
Solution Approach 2:
The patent introduces an intermediary geometric model that bridges the gap between 2D image coordinates and 3D gaze point determination. This intermediary model serves as a mediator that transforms face feature points into a 3D coordinate system, enabling gaze point calculation without losing the underlying geometric relationships that deep learning might otherwise discard.
2Device complexity
If only two-dimensional coordinates of eye feature points are used, then the processing is simple, but the gaze point determination accuracy is insufficient
Solution Approach 1:
The patent transitions from 2D image coordinates to a 3D coordinate system by constructing a three-dimensional face model. This dimensionality change allows the system to capture depth information and spatial relationships that are lost in 2D, thereby improving gaze point determination accuracy while adding only moderate computational complexity through geometric transformations.
Data Source
AI summary
A gaze point determination method and apparatus, an electronic device, and a computer storage medium are provided. The method includes: obtaining two-dimensional coordinates of eye feature points of at least one eye of a face in an image, the eye feature points including an eyeball center area feature point; obtaining, in the preset three-dimensional coordinate system, three-dimensional coordinate of a corresponding eyeball center area feature point in a three-dimensional face model corresponding to the face in the image based on the obtained two-dimensional coordinate of the eyeball center area feature point; and obtaining a determination result for a position of a gaze point of the eye of the face in the image according to two-dimensional coordinates of feature points other than the eyeball center area feature point in the eye feature points and the three-dimensional coordinate of the eyeball center area feature point in the preset three-dimensional coordinate system.