Image-Fitted 3D Face Models for Large-Angle Pose Estimation

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Existing face pose estimation methods using three-dimensional average face models suffer from reduced accuracy when dealing with large-angle and exaggerated facial expressions.

Innovation Solution

Construct a three-dimensional face model corresponding to a two-dimensional face image using a face shape and/or facial expression fitting algorithm, incorporating three-dimensional face prior data to align and fit feature points, thereby improving robustness and accuracy of pose estimation.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Ease of manufacture

If a three-dimensional average face model is used for pose estimation, then the method is simple and computationally efficient, but the accuracy deteriorates for large-angle and exaggerated facial expressions

Engineering Contradiction:
Improvesimplicity of pose estimation methodVSAvoidpose estimation accuracy
Core Design Contradiction:
Ease of manufactureVSMeasurement precision

Solution Approach 1:

The patent applies dynamics by transitioning from a static three-dimensional average face model to a dynamic three-dimensional face model that can adapt to various facial expressions and poses. The system reconstructs a three-dimensional face model from a two-dimensional face image, allowing the model to dynamically adjust to different facial configurations rather than relying on a fixed average model, thereby maintaining high accuracy across diverse facial conditions.

Inventive Principle:
Principle #15Dynamics

Solution Approach 2:

The patent utilizes parameter changes by transforming the face representation from a fixed three-dimensional average model to a parameterized three-dimensional model that can be adjusted based on the specific two-dimensional image input. By extracting feature points and reconstructing the face model with adjustable parameters, the system can adapt to different facial geometries and expressions, resolving the contradiction between model simplicity and estimation accuracy.

Inventive Principle:
Principle #35Parameter changes

2Loss of time

If a three-dimensional average face model is used, then the computation time is short, but the robustness deteriorates for diverse facial expressions and poses

Engineering Contradiction:
Improvecomputation timeVSAvoidrobustness for diverse expressions
Core Design Contradiction:
Loss of timeVSReliability

Solution Approach 1:

The patent applies preliminary action by pre-establishing a three-dimensional face model framework that can be quickly instantiated and adjusted for different inputs. The system pre-processes the two-dimensional image to extract key feature points and then efficiently reconstructs the corresponding three-dimensional model using the pre-defined framework, reducing computation time while maintaining robustness for diverse facial expressions and poses.

Inventive Principle:
Principle #10Preliminary action

Solution Approach 2:

The patent utilizes copying by creating a three-dimensional copy of the face from the two-dimensional image input. Instead of using a fixed average model, the system generates a personalized three-dimensional representation that mirrors the specific facial geometry in the input image, enabling accurate pose estimation for diverse expressions while maintaining computational efficiency through optimized reconstruction algorithms.

Inventive Principle:
Principle #26Copying

3Measurement precision

If the three-dimensional model is constructed to match the two-dimensional image accurately, then the pose estimation accuracy improves, but the device complexity increases

Engineering Contradiction:
Improvepose estimation accuracyVSAvoidcomplexity of face model construction
Core Design Contradiction:
Measurement precisionVSDevice complexity

Solution Approach 1:

The patent applies segmentation by dividing the face model construction process into distinct segments: extracting feature points from the two-dimensional image, reconstructing the three-dimensional face model based on these points, and determining pose parameters. This segmentation allows each step to be optimized independently, maintaining high accuracy while managing complexity through modular processing stages.

Inventive Principle:
Principle #1Segmentation

Solution Approach 2:

The patent utilizes an intermediary approach by introducing a three-dimensional face model as an intermediate representation between the two-dimensional image and the final pose estimation. This intermediary model serves as a bridge that translates two-dimensional image data into three-dimensional spatial understanding, enabling accurate pose estimation without directly complexifying the overall system architecture.

Inventive Principle:
Principle #24Intermediary (Mediator)

Data Source

PatentEP3836070B1Face pose estimation/three-dimensional face reconstruction method and apparatus, and electronic device
Publication Date: 2025.10.15 ALIBABA GROUP HOLDING LTD
  • EP3836070B1 patent drawingFigure 1
  • EP3836070B1 patent drawingFigure 2
  • EP3836070B1 patent drawingFigure 3

AI summary

This application discloses methods, apparatus, and electronic devices for face pose estimation and three-dimensional face reconstruction. The face pose estimation method comprises: acquiring a two-dimensional face image for processing, constructing a three-dimensional face model corresponding to the two-dimensional face image, and determining a face pose of the two-dimensional face image based on face feature points of the three-dimensional face model and face feature points of the two-dimensional face image. With this approach, the face pose estimation is performed based on the three-dimensional face model corresponding to the two-dimensional face image, instead of only based on a three-dimensional average face model. As a result, a high accuracy pose estimation can be obtained even for a face with large-angle and exaggerated facial expressions. Thus, robustness of the pose estimation can be effectively improved.