Face Location Detection via 3D Surface Angle Data
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing methods for face location determination in 3D models are not robust against variations in face rotation and rely on unreliable optical surface properties, making them person-dependent and inefficient.
Innovation Solution
A system and method using a 3D surface model derived from 2D optical imaging to generate a 2D angle data image, which is then processed with a 2D face location algorithm, allowing for robust face location detection independent of optical properties and person-specific features, using multiple virtual lighting directions for improved accuracy.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If 2D optical images are used for face location determination, then the method is simpler and faster, but it is not robust against variations in face rotation and optical surface properties
Solution Approach 1:
The patent transforms the face location detection problem from 2D image space to 3D surface model space by generating a 3D surface model from 2D optical images. This dimensional transformation enables the system to capture geometric information that is invariant to lighting conditions and face rotation, thereby improving robustness while maintaining the efficiency of 2D image acquisition
2Reliability
If 3D surface models are used for face location determination, then robustness against optical surface properties is improved, but the processing complexity increases
Solution Approach 1:
The patent extracts only the essential geometric information from the 3D surface model by generating a 2D angle data image that represents angles between normals to the modeled 3D surface and incidence directions. This extraction approach maintains the robustness benefits of 3D modeling while reducing processing complexity by working with a simplified 2D representation derived from the 3D model
3Measurement precision
If person-specific features are used for face location, then face recognition accuracy is improved, but the method becomes less generalizable across different persons
Solution Approach 1:
The patent creates a normalized angular representation of the 3D surface geometry that captures the essential shape characteristics of the face without encoding person-specific details. This angular copy of the surface geometry can be used for location determination across different persons, achieving both accuracy and person-independence
Data Source
AI summary
The location of a face is detected from data about a scene. A 3D surface model from is obtained from measurements of the scene. A 2D angle data image is generated from the 3D surface model. The angle data image is generated for a virtual lighting direction, the image representing angles between a ray directions from a virtual light source direction and normal to the 3D surface. A 2D face location algorithm is applied to each of the respective 2D images. In an embodiment respective 2D angle data images for a plurality of virtual lighting directions are generated and face locations detected from the respective 2D images are fused.