Two-Stage Face Recognition for False Detection Filtering
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing face recognition systems struggle with inaccurate and time-consuming face detection due to varying background conditions, leading to erroneous detections and reduced response speed, especially in uncontrolled environments like entrance halls.
Innovation Solution
A dual-threshold face recognition system that uses a low-accuracy, high-speed initial detection followed by high-accuracy verification, combined with automatic learning to adjust for erroneous detections, allowing for rapid and precise face recognition without requiring strict positioning or background constraints.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If a single high-accuracy threshold is used for face detection, then detection accuracy is improved, but processing speed deteriorates and false positives increase in varying background conditions
Solution Approach 1:
The face detection process is segmented into two distinct stages: a first face recognition application performs initial detection with a first threshold value, and a second face recognition application performs verification with a second threshold value. This segmentation allows the system to balance speed and accuracy by processing most images quickly while only subjecting potentially positive cases to the more computationally intensive second stage.
Solution Approach 2:
The system dynamically adjusts the threshold value based on the processing stage. The first threshold value is set lower for rapid initial screening, while the second threshold value is set higher for accurate verification. This dynamic adjustment of parameters allows the system to adapt its sensitivity and processing intensity to the specific needs of each detection stage.
2Measurement precision
If a single high-accuracy threshold is used for face detection, then detection accuracy is improved, but the number of false positives increases due to background variations
Solution Approach 1:
The verification process is segmented into two applications with different threshold characteristics. The first application uses a more sensitive threshold to catch all potential faces, while the second application uses a stricter threshold to eliminate false positives. This two-stage segmentation effectively filters out background variations that would trigger false positives in a single-threshold system.
Solution Approach 2:
The first face recognition application acts as an intermediary between the captured image and the second face recognition application. It performs preliminary filtering to identify candidate regions, reducing the burden on the second application and allowing it to focus computational resources on verifying only the most promising candidates, thereby reducing false positives.
3Device complexity
If conventional face detection is used in uncontrolled environments, then system simplicity is maintained, but detection accuracy deteriorates due to background noise
Solution Approach 1:
The detection system is segmented into two functional layers: a fast initial detection layer that maintains simplicity and a verification layer that enhances accuracy. The first face recognition application provides the simple, rapid initial screening, while the second application adds the accuracy enhancement needed for uncontrolled environments without requiring complex hardware modifications.
Solution Approach 2:
The system changes the threshold parameter between the two detection stages to adapt to varying background conditions. The first threshold is optimized for speed and broad coverage, while the second threshold is optimized for accuracy in challenging environments. This parameter change allows the system to maintain simplicity in structure while achieving high accuracy through adaptive parameter selection.
Data Source
Figure 1
Figure 2
Figure 3
AI summary
A face recognition technology that enables high-speed determination of erroneous detection of face regions and high-precision face recognition by means of a first face recognition processing unit of relatively low accuracy for detecting a face region of a person on each captured image captured by a monitor camera, and a second face recognition processing unit for performing face recognition of a person more accurately with respect to the detected face region transmitted from the first face recognition processing unit. The second face recognition processing unit determines that the first face recognition processing unit has erroneously detected when the second face recognition processing unit is not able to recognize a face region on the detected face region transmitted from the first face recognition processing unit, and the first face recognition processing unit calculates and self-learns a barycentric coordinate of the detected face region as a point mask position. Thereafter, the first face recognition processing unit checks whether or not the coordinate of the barycentric position has already been stored in a storage as one of the point mask position, and, when the coincident coordinate of the point mask position is found in the data of the barycentric coordinates, this face recognition system does not perform more accurate face recognition processing by the second face recognition processing unit, and proceeds to the next face recognition processing on the next captured image.