3D Face Model Generation for Pose-Invariant Recognition

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Existing facial recognition systems face challenges in identifying individuals from images captured under unconstrained conditions due to variations in lighting, viewing angles, occlusions, and age, which complicates the process of sifting through large amounts of image data from diverse sources.

Innovation Solution

A method and system that determines a three-dimensional (3D) model of a face from multiple images, extracts two-dimensional (2D) patches, and generates multi-view probabilistic elastic parts (PEP) signatures, accounting for geometric and photometric variability, and incorporating attributes like gender, age, and ethnicity, to create an attribute-based representation that is pose-invariant and resilient to occlusions and low-resolution data.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Measurement precision

If traditional 2D facial recognition methods are used on images from unconstrained conditions, then the system can process images quickly, but the recognition accuracy deteriorates due to variations in lighting, viewing angles, occlusions, and age

Engineering Contradiction:
Improvefacial recognition accuracyVSAvoidrobustness to unconstrained conditions
Core Design Contradiction:
Measurement precisionVSAdaptability or versatility

Solution Approach 1:

The patent transforms 2D facial images into a 3D model representation, adding a dimensional aspect to handle variations in viewing angle and lighting. The 3D model allows the system to normalize facial features across different poses and lighting conditions, thereby improving recognition accuracy in unconstrained conditions without sacrificing adaptability

Inventive Principle:
Principle #17Another dimension (Dimensionality change)

Solution Approach 2:

The system changes the parameter space from 2D image coordinates to 3D geometric parameters and appearance parameters. By representing faces in 3D space with parameters for shape, texture, and lighting, the system can normalize variations caused by different viewing angles and lighting conditions, resolving the contradiction between accuracy and adaptability

Inventive Principle:
Principle #35Parameter changes

2Reliability

If multiple images are processed to create comprehensive facial representations, then recognition robustness improves, but the time and computational resources required increase

Engineering Contradiction:
Improverecognition robustnessVSAvoidprocessing time
Core Design Contradiction:
ReliabilityVSLoss of time

Solution Approach 1:

The patent performs preliminary actions by constructing a 3D model from multiple images in advance, organizing facial data into a structured representation that captures geometric and appearance variations. This pre-processing creates a compact facial signature that can be quickly compared during recognition, thereby maintaining robustness while reducing real-time processing time

Inventive Principle:
Principle #10Preliminary action

Solution Approach 2:

The system creates a simplified 3D model copy that captures the essential facial characteristics without storing all original images. This compact representation serves as a surrogate for the full set of images, maintaining recognition reliability while significantly reducing processing time and computational resources

Inventive Principle:
Principle #26Copying

3Adaptability or versatility

If 3D models are constructed from multiple images to handle pose variations, then pose invariance improves, but the system complexity increases

Engineering Contradiction:
Improvepose invarianceVSAvoidsystem complexity
Core Design Contradiction:
Adaptability or versatilityVSDevice complexity

Solution Approach 1:

The patent segments the facial recognition system into distinct modules: image processing to extract facial features, 3D model construction to represent geometric structure, and signature generation for comparison. This segmentation allows each component to handle specific aspects of pose variation independently, achieving pose invariance while managing system complexity through modular design

Inventive Principle:
Principle #1Segmentation

Data Source

PatentEP3084682B1System and method for identifying faces in unconstrained media
Publication Date: 2019.07.24 AVIGILON FORTRESS
  • EP3084682B1 patent drawingFigure 1
  • EP3084682B1 patent drawingFigure 2
  • EP3084682B1 patent drawingFigure 3

AI summary

Methods and systems for facial recognition are provided. The method includes determining a three-dimensional (3D) model of a face of an individual based on different images of the individual. The method also includes extracting two-dimensional (2D) patches from the 3D model. Further, the method includes generating a plurality of signatures of the face using different combinations of the 2D patches, wherein the plurality of signatures correspond to respective views of the 3D model from different angles.