Multi-Person Video Tracking With Quality-Scored Face Crops
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing video surveillance systems struggle to perform facial recognition for multiple individuals in video data in real-time, often failing to generate high-quality cropped images for efficient identity data processing.
Innovation Solution
A compute device within a video camera system tracks multiple individuals using motion models, generates cropped images based on quality scores and elapsed time, and prioritizes facial recognition tasks to optimize processing efficiency.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If facial recognition is performed for multiple individuals in real-time, then recognition speed and productivity improve, but image quality and processing accuracy deteriorate due to resource constraints
Solution Approach 1:
The system segments the video processing task by generating and processing cropped images of individual faces separately from the full video stream. Each face detection creates a dedicated cropped image that can be processed independently, allowing multiple individuals to be recognized simultaneously without compromising the quality of individual face analysis.
Solution Approach 2:
The system extracts only the relevant portions (faces) from the full video stream by generating cropped images based on detected face locations. This extraction allows the system to focus computational resources on the most important elements (faces) rather than processing entire video frames, thereby improving recognition speed while maintaining image quality for the extracted portions.
2Measurement precision
If cropped images are generated for all detected individuals, then recognition accuracy improves, but processing time and computational resources increase
Solution Approach 1:
The system applies local quality by generating cropped images only for detected faces that meet specific quality thresholds. Instead of processing all detected individuals uniformly, the system prioritizes faces with sufficient image quality for accurate recognition, thereby reducing processing time for low-quality detections while maintaining high accuracy for quality-validated faces.
Solution Approach 2:
The system performs preliminary actions by generating quality scores for cropped images before final recognition processing. This preliminary quality assessment allows the system to filter out low-quality detections early in the process, avoiding unnecessary computational resources for faces that cannot be reliably recognized, thus reducing overall processing time while maintaining accuracy for valid faces.
3Productivity
If quality scores are assigned to cropped images, then processing efficiency improves, but system complexity increases
Solution Approach 1:
The system implements self-service by using the cropped images themselves as the basis for generating quality scores. The quality assessment is performed automatically on the image data without requiring external intervention or complex additional hardware. This self-service approach improves processing efficiency through automated quality filtering while keeping the added complexity manageable by using the existing image processing infrastructure.
Data Source
AI summary
A non-transitory, processor-readable medium stores instructions that, when executed by a processor, cause the processor to receive video-derived detection data associated with a plurality of persons. For a first person from the plurality of persons, a portion of the video-derived detection data associated with the first person is assigned to a first motion track based on a motion model, and a closeup image of the first person is generated based on the portion of video-derived detection data. A quality score is generated based on the closeup image, and the closeup image is assigned to the first motion track based on the quality score. The first motion track is selected from a plurality of motion tracks associated with the plurality of persons. Using a neural network, first identity data is generated based on the closeup image, and the first motion track is updated based on the first identity data.