Face Pose Estimation via Statistical-Structural Fusion

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Current face detection methods face challenges with high computation requirements, limited performance, especially for multi-view face detection, and difficulties in handling in-plane and out-of-plane rotations, leading to inefficiencies and inaccuracies in face pose estimation.

Innovation Solution

A fusion of statistical and structural face detection methods is employed, where coarse pose information is determined using a statistical method, followed by fine pose determination using a structural method, and the image is rotated to verticality for accurate face detection, reducing the need for multiple detectors and improving time efficiency.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Adaptability or versatility

If statistical methods are used for face detection by scanning images on multiple scales, then face detection coverage is improved, but computation capability requirements increase significantly

Engineering Contradiction:
Improveface detection coverageVSAvoidcomputation capability
Core Design Contradiction:
Adaptability or versatilityVSPower

Solution Approach 1:

The face detection process is segmented into two distinct stages: a detection stage using a detector trained on frontal faces, and a pose estimation stage using a separate pose estimator. This segmentation allows the system to first identify potential face locations efficiently, then estimate pose only for those candidates, reducing overall computational burden while maintaining comprehensive detection coverage across multiple scales and orientations.

Inventive Principle:
Principle #1Segmentation

Solution Approach 2:

The detector performs preliminary face candidate identification before pose estimation is conducted. By pre-processing the image to locate potential faces using a computationally efficient detector, the system avoids applying heavy pose estimation algorithms to entire images, thereby reducing power consumption while ensuring comprehensive face detection coverage.

Inventive Principle:
Principle #10Preliminary action

2Power

If structure-based methods are used for face detection, then computation requirements are reduced, but false alarm rate increases

Engineering Contradiction:
Improvecomputation requirementsVSAvoidfalse alarm rate
Core Design Contradiction:
PowerVSReliability

Solution Approach 1:

The system merges two complementary approaches: a detector based on statistical methods (e.g., Haar features with AdaBoost) that provides high reliability with controlled false alarms, and a structure-based pose estimator that uses facial landmarks and geometric constraints. The detector's reliable face candidate selection combined with the pose estimator's structural verification creates a hybrid system that maintains low false alarm rates while reducing overall computation requirements.

Inventive Principle:
Principle #5Merging (Combining)

3Adaptability or versatility

If multiple detectors are used to handle different face orientations, then multi-view face detection performance is improved, but device complexity increases

Engineering Contradiction:
Improvemulti-view face detection performanceVSAvoidnumber of detectors
Core Design Contradiction:
Adaptability or versatilityVSDevice complexity

Solution Approach 1:

The pose estimator is designed with multi-functionality to handle various face orientations and rotations. Instead of deploying separate specialized detectors for each view, a single universal pose estimator processes all detected face candidates, estimating pose parameters that accommodate frontal, profile, and intermediate orientations. This universal approach maintains multi-view detection performance while significantly reducing device complexity.

Inventive Principle:
Principle #6Universality (Multi-functionality)

4Measurement precision

If pose estimation is performed for all face candidates, then accurate face detection is improved, but processing time increases

Engineering Contradiction:
Improveface detection accuracyVSAvoidprocessing time
Core Design Contradiction:
Measurement precisionVSLoss of time

Solution Approach 1:

The system applies pose estimation selectively rather than universally - performing the computationally intensive pose estimation only on face candidates that meet certain confidence thresholds or exhibit specific characteristics. This partial action approach maintains accurate face detection for critical cases while avoiding unnecessary processing time for obvious or low-priority candidates, thereby balancing precision with processing efficiency.

Inventive Principle:
Principle #16Partial or excessive action

Data Source

PatentEP2370932B1Method, apparatus and computer program product for providing face pose estimation
Publication Date: 2018.04.04 NOKIA TECHNOLOGIES OY
  • EP2370932B1 patent drawingFigure 1
  • EP2370932B1 patent drawingFigure 2
  • EP2370932B1 patent drawingFigure 3

AI summary

A method for providing face pose estimation for face detection may include utilizing a selected portion of classifiers in detectors to determine coarse pose information for a candidate face in an image, determining fine pose information for the candidate face based at least in part on the determined coarse pose information, and employing another portion of the classifiers in the detectors to perform face detection based at least in part on the fine pose information to determine whether the candidate face corresponds to a face. An apparatus and computer program product corresponding to the method are also provided.