Face Pose Detection Reliability Scoring
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing face detection systems face challenges in accurately determining the pose of faces in images, particularly when multiple poses are detected from the same face, leading to low detection accuracy and slow processing speeds, and often produce multiple incorrect detection results.
Innovation Solution
An image processing apparatus that includes a detection section to identify face poses and a determination section to calculate the reliability of detection results, comparing them to determine the correct pose by adjusting detection window settings and using correction coefficients, thereby prioritizing detection results based on the input image type (still or moving).
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If multiple face detectors (frontal, profile, etc.) are used to detect faces in various poses, then detection coverage is improved, but multiple detection results are generated from one face leading to determination difficulty
Solution Approach 1:
The determination section uses feedback mechanisms by calculating reliability scores for each detection result and using this information to select the most accurate pose. The system continuously refines its determination by comparing detection results across multiple detectors and using reliability feedback to resolve conflicts between overlapping detections.
Solution Approach 2:
The determination section acts as an intermediary between multiple face detectors and the final output. It receives detection results from various detectors, calculates reliabilities, and mediates the selection process by choosing the pose with the highest reliability, thus resolving the conflict between multiple detection results.
2Adaptability or versatility
If a single face detector learns all poses from enormous sample data, then all poses can be detected, but detection accuracy and speed decrease
Solution Approach 1:
The face detection system is segmented into multiple specialized detectors, each trained on specific pose types (frontal, profile, etc.). This segmentation allows each detector to achieve high accuracy for its specific pose type while maintaining comprehensive coverage through the combination of multiple detectors.
Solution Approach 2:
The determination section provides universal functionality by handling the selection of correct poses across all detection types. It serves as a multi-functional component that works with any combination of detectors and determines the correct pose regardless of which detector produced the detection.
3Adaptability or versatility
If a single face detector learns all poses from enormous sample data, then all poses can be detected, but detection speed decreases
Solution Approach 1:
By segmenting the detection task into multiple specialized detectors, each detector can be optimized for its specific pose type with fewer training samples, reducing the computational burden per detector and improving overall detection speed while maintaining comprehensive pose coverage.
Solution Approach 2:
Each detector performs partial action by focusing on specific pose types rather than attempting to detect all poses. This partial specialization allows for faster, more efficient detection of each pose type, and the determination section combines these partial results to achieve complete pose detection.
4Adaptability or versatility
If multiple detection results are generated from one face, then all poses are captured, but it becomes problematic to determine which detection result is correct
Solution Approach 1:
The determination section implements feedback by calculating reliability scores for each detection result based on multiple factors including detection confidence, pose overlap analysis, and consistency with other detectors. This reliability feedback enables the system to identify and select the correct pose from multiple detection results.
Solution Approach 2:
The system changes parameters by introducing reliability scores as a new parameter to evaluate detection results. By calculating and comparing reliability parameters across multiple detection results, the system can objectively determine which pose detection is correct, transforming the subjective problem of pose selection into an objective parameter-based decision.
Data Source
AI summary
An electronic image is received by a system to process the image for the presence of a face. The image is repeatedly electronically scanned using a plurality of windows for the presence of facial poses. A plurality of directional poses is detected during the scanning process. Reliabilities for each type of detected poses are calculated. The reliabilities are based on the amount of times the directional poses are detected during the scanning process and directions of the directional poses.