Facial Recognition Accuracy Through Detector-Tracker Face Selection
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Facial recognition systems face challenges in accurately identifying individuals due to noise, illumination, motion blur, and imperfections in computer-based recognition functions, particularly in public events and venues.
Innovation Solution
A method and system that utilizes a detector and tracker to identify and associate faces without a facial recognizer, applying selected faces to a facial classifier after adjusting thresholds based on scores, and employing edge computing devices to process video streams from multiple cameras, enhancing accuracy through tracking and normalization.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If facial recognition is applied to every detected face in real-time, then identification accuracy may improve, but computational resource consumption and processing time increase significantly
Solution Approach 1:
The system performs preliminary actions by detecting faces and tracking them across frames before applying facial recognition. Faces are pre-processed and scored based on detection confidence and tracking consistency, so that only high-quality candidate faces are submitted for resource-intensive recognition analysis. This preliminary filtering prepares the data in advance to reduce unnecessary computational overhead during recognition.
Solution Approach 2:
Instead of applying facial recognition to every detected face, the system applies it selectively to only those faces that meet certain confidence thresholds and tracking criteria. This partial action approach applies recognition where most needed (high-confidence tracked faces) while avoiding waste on low-quality detections, thereby balancing accuracy improvement with resource conservation.
2Reliability
If facial recognition is applied to every detected face, then comprehensive identification may be achieved, but false positives increase due to noise, illumination, and motion blur
Solution Approach 1:
The system performs preliminary face detection and tracking to establish confidence scores before recognition. By pre-assessing image quality metrics (detection confidence, tracking consistency) and preparing only high-quality candidates for recognition, the system reduces the impact of noise, illumination variations, and motion blur that would otherwise cause false positives.
Solution Approach 2:
The system uses feedback from face detection confidence scores and tracking consistency to dynamically adjust which faces are submitted for recognition. High-confidence tracked faces receive recognition analysis, while low-confidence detections are filtered out. This feedback mechanism continuously optimizes the balance between comprehensive identification and false positive reduction.
3Productivity
If multiple faces are processed simultaneously, then throughput increases, but resource consumption and complexity of managing multiple tracks increase
Solution Approach 1:
The system segments the processing workflow into distinct stages: face detection, tracking across frames, scoring/filtering, and recognition. Each stage handles a specific subset of faces independently, managing complexity through modular segmentation. Multiple faces are processed through these segmented stages in parallel, increasing throughput while keeping each stage's complexity manageable.
Solution Approach 2:
The tracking module serves multiple functions simultaneously: it maintains face tracks across frames, scores detection confidence, filters low-quality candidates, and prepares data for recognition. This multi-functionality allows the system to process multiple faces efficiently through a single unified tracking system rather than requiring separate processing paths for each function.
Data Source
AI summary
The invention provides, in some respects, a digital data processing method of identifying a person of interest in a video stream. The method, which executes on a digital data processing system including one or more digital data processing devices in communications coupling with one another, includes detecting using a detector and without using a facial recognizer one or more faces in a video stream. It further includes using a tracker to associate, without using a facial recognizer, one or more of the detected faces with a same apparent individual, and scoring, again, without using a facial recognizer, each of the detected faces that are associated with the same apparent individual. The method further includes applying selected ones of the scored detected faces of the same apparent individual to a facial recognizer to identify a person of interest, where selection is based on scores of the detected faces.