Normalized Virtual Camera for Facial Feature Detection
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing methods for extracting feature information from human faces in input images, such as cropping and scaling followed by orthographic projection, fail to preserve perspective projection-induced distortions, leading to increased inference errors, especially when capturing images at close range or with significant face rotations.
Innovation Solution
The proposed solution involves setting a normalized virtual camera that applies a perspective projection model to the input image, generating a normalized image that preserves perspective projection features, and then inputting this image to an AI model trained to extract facial feature information. This approach transforms the feature information into the camera's coordinate system using spatial transformation matrices between cameras.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If cropping and scaling methods are used for image normalization, then the input image can be processed to match AI model requirements, but perspective projection-induced distortions are lost leading to increased inference errors
Solution Approach 1:
The patent changes the projection parameter from orthographic to perspective projection model. By using perspective projection for image normalization instead of traditional orthographic projection, the system preserves perspective projection-induced distortions that are crucial for accurate facial feature detection, especially in close-range imaging scenarios.
Solution Approach 2:
The patent creates a normalized virtual camera that copies the perspective projection characteristics of the actual camera. This virtual camera system replicates the perspective distortion effects in a controlled normalization process, allowing the AI model to learn and process facial features while maintaining the perspective information needed for accurate inference.
2Device complexity
If orthographic projection-based AI models are used, then the model structure is simpler, but inference errors increase when capturing images at close range or with significant face rotations
Solution Approach 1:
The patent applies preliminary action by normalizing the input image using perspective projection before feeding it to the AI model. This pre-processing step transforms the image into a normalized coordinate system that preserves perspective distortions, enabling the model to handle close-range and rotated faces more accurately without changing the fundamental model architecture.
Solution Approach 2:
The patent introduces a normalized virtual camera as an intermediary between the actual camera and the AI model. This virtual camera system acts as a mediator that translates real-world images into a normalized representation space, bridging the gap between simple model architecture and complex real-world imaging conditions.
3Reliability
If perspective projection model is applied for normalization, then perspective projection-induced distortions are preserved reducing inference errors, but the normalization process becomes more complex
Solution Approach 1:
The patent achieves universality by creating a normalized virtual camera system that can handle multiple imaging scenarios (close-range, rotated faces, different poses) through a single unified perspective projection-based normalization approach. This multi-functional system addresses various real-world conditions without requiring separate processing pipelines for each scenario.
Solution Approach 2:
The patent applies dimensionality change by transitioning from 2D orthographic projection to 3D perspective projection in the normalization process. This dimensional transformation preserves the depth information and perspective distortions that are essential for accurate facial feature detection, adding a third dimension to the normalization operation.
Data Source
AI summary
An electronic device for detecting feature information from a face and an operating method of the electronic device are disclosed. The operating method may include: acquiring, via a camera, an input image including a person; setting a normalized virtual camera for generating a normalized image from the input image; generating a normalized image including a perspective projection feature that includes a face of the person based on the normalized virtual camera; inputting the normalized image to an artificial intelligence (AI) model trained to extract feature information of the face; and transforming the feature information of the face output from the AI model into a first coordinate system which is a coordinate system of the camera, using a spatial transformation relationship between the camera and the normalized virtual camera.