Face Pose Registration via Dynamic Angle Estimation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional face recognition methods are inadequate in handling pose variations, as they either require predefined poses during registration, leading to poor performance with new poses, or rely on precise 3D modeling, which is time-consuming and prone to errors in shape representation.
Innovation Solution
An apparatus and method that automatically registers face information based on various poses by detecting and recognizing face poses using a face detection unit and pose recognition unit, providing an interface for registering unregistered poses and comparing them to existing registered information for similarity.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Ease of operation
If multiple side profile images are registered and used to recognize faces, then face recognition can be performed with predefined poses, but face recognition performance deteriorates when images of faces in new poses are input
Solution Approach 1:
The patent applies dynamics by transitioning from static predefined pose templates to dynamic pose estimation. The system continuously estimates face pose angles (pitch, yaw, roll) from input images and dynamically selects appropriate registered face images based on the estimated pose, enabling adaptation to new poses without requiring predefined templates for each pose angle.
Solution Approach 2:
The patent utilizes parameter changes by representing face poses as continuous angular parameters (pitch, yaw, roll angles) rather than discrete predefined poses. This allows the system to handle a continuous range of poses by registering face images at multiple pose angles and selecting based on the estimated angular parameters, thereby improving versatility while maintaining ease of operation.
2Adaptability or versatility
If a 3D face model is generated from face images to handle pose variations, then face recognition becomes robust to pose changes, but the process becomes time-consuming and may produce inaccurate face shapes
Solution Approach 1:
The patent extracts the pose information from the 3D face model generation process and uses it directly for face image selection. Instead of generating a complete 3D model and then extracting pose data, the system directly estimates pose angles from 2D face images and uses these extracted pose parameters to select appropriate registered face images, thereby reducing time consumption while maintaining pose variation robustness.
Solution Approach 2:
The patent applies copying by creating a database of registered face images at multiple pose angles as copies of the same face. During recognition, the system selects the most similar copy (registered face image) based on the estimated pose angle, avoiding the need for time-consuming 3D model generation while achieving robustness to pose variations through these pose-specific copies.
3Measurement precision
If a 3D face model is generated using precise 3D sensors, then face shape accuracy is improved, but the complexity and cost of the system increases
Solution Approach 1:
The patent employs the principle of using simple, readily available 2D camera images instead of expensive 3D sensors. By registering face images at multiple pose angles captured by standard cameras and using pose estimation algorithms, the system achieves sufficient face shape and pose information without requiring complex and costly 3D sensing hardware, thereby reducing device complexity while maintaining measurement precision for recognition purposes.
4Measurement precision
If the number of registered face images is increased to improve recognition accuracy, then face recognition becomes more accurate, but the time taken to recognize a face increases
Solution Approach 1:
The patent applies segmentation by dividing the face recognition task into pose-based segments. Instead of comparing the input image against all registered face images uniformly, the system first estimates the face pose angle and then selects only the registered face images that correspond to similar pose angles. This segmentation based on pose categories significantly reduces the number of images to be compared, thereby maintaining high recognition accuracy while improving recognition speed.
Data Source
AI summary
Disclosed herein are an apparatus and method for registering face poses for face recognition. The apparatus includes a face detection unit for detecting the face of a user from an image including the face of the user; a pose recognition unit for recognizing the face pose of the face of the user based on the degree of rotation of the face of the user; a registration interface unit for providing an interface for showing information about whether the face poses of the face of the user are registered; and a face registration unit for registering the face pose of the face of the user when the face pose of the face of the user is recognized as an unregistered face pose based on the interface.