3D Face Model Adaptation for Accessory-Aware Recognition
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing facial recognition devices struggle to accurately detect and recognize faces when individuals wear accessories such as sunglasses, masks, or hats, especially when the face is oriented sideways or partially hidden by collars or reflections are absent.
Innovation Solution
A facial recognition device that includes a determination section to identify accessories, an acquisition section to gather height information, a production section to create a three-dimensional face model adapted for the accessory, and a fitting section to align the face model with the captured image, allowing for accurate recognition even with accessories present.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If conventional facial recognition methods are used, then the device can detect faces under normal conditions, but it fails to accurately detect faces when accessories are worn or face orientation changes
Solution Approach 1:
The patent applies dynamics by making the 3D face model adaptable and adjustable based on detected accessory types. The system dynamically modifies the base 3D face model by adding accessory components (sunglasses, hats, masks) to create customized models that match the actual facial configuration, enabling accurate recognition despite accessory variations and different face orientations.
2Reliability
If background reflection detection is used for eyeglass detection, then eyeglasses can be detected when reflection is present, but face orientation detection becomes impossible when reflection is absent
Solution Approach 1:
The patent uses an intermediary approach by introducing a comprehensive 3D face model that incorporates accessory geometry as a mediator between the image data and the detection algorithm. Instead of relying solely on reflection patterns or accessory-specific features, the system uses the structured 3D model to provide geometric constraints and expected feature locations, enabling detection across various lighting conditions and orientations.
3Productivity
If standard 2D image processing is used, then processing speed is maintained, but accurate recognition is impossible when face parts are hidden by accessories or collars
Solution Approach 1:
The patent transitions from 2D image processing to 3D face modeling, adding a spatial dimension to the recognition process. By constructing and matching 3D face models that incorporate accessory geometry, the system can reason about hidden features and infer their positions based on the 3D structure, maintaining processing efficiency while significantly improving detection precision in challenging conditions.
Data Source
AI summary
A facial recognition device includes a determination section for determining presence of an accessory worn on a face from a captured image capturing a human face, an acquisition section for acquiring height information indicative of a height of the accessory from the surface of the face in case the determination section determines presence of the accessory on the face, a production section for producing a three-dimensional face model based on the height information and a fitting section for fitting the face included in the captured image, with using the three-dimensional face model.