Face Detection Pipeline for Real-Time Crowd Censusing
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing facial recognition systems and object recognition systems are costly and inefficient for real-time crowd monitoring and counting in large gatherings, as they require expensive hardware and processing resources, limiting their utility in crowd control situations.
Innovation Solution
A face detection pipeline-based method using dual people detectors with differing capabilities to estimate and correct crowd statistics, including the number of individuals viewing content and their duration of attention, allowing for real-time monitoring and content adaptation without the need for expensive hardware.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If expensive facial recognition systems are used to detect and recognize faces in crowd scenes, then measurement precision of individual identification is improved, but device complexity and processing resource requirements worsen
Solution Approach 1:
The system segments the face detection task into multiple independent detectors with different capabilities. A first people detector provides basic face detection, while a second people detector provides enhanced detection with better accuracy. This segmentation allows the system to achieve high measurement precision for crowd counting without deploying expensive high-performance detectors throughout the entire system, thus reducing overall device complexity and processing resource requirements.
Solution Approach 2:
The patent applies local quality by using different detector types for different functional requirements. The first people detector is optimized for speed and basic detection in regions where rapid counting is critical, while the second people detector with higher precision is applied selectively in regions or time periods where accurate measurement is most important. This localized application of quality optimizes the balance between measurement precision and device complexity.
2Measurement precision
If advanced body recognition systems are used to detect movement and track individuals, then measurement precision of crowd activity detection is improved, but device complexity and processing resource requirements worsen
Solution Approach 1:
The system applies partial action by using the second people detector (with higher computational cost) only for a subset of frames or time periods rather than continuously processing all video frames. This selective application maintains measurement precision for detecting crowd activity trends while significantly reducing overall processing resource requirements and device complexity.
Solution Approach 2:
The first people detector performs preliminary face detection and counting on all frames to establish baseline crowd statistics. Only after this preliminary action does the system apply the more resource-intensive second people detector to analyze specific frames for movement detection and activity tracking. This preliminary action filters out unnecessary processing, reducing overall device complexity while maintaining measurement precision.
3Productivity
If real-time crowd monitoring is implemented using comprehensive detection methods, then productivity of crowd analysis is improved, but device complexity and processing resource requirements worsen
Solution Approach 1:
The patent merges the outputs of multiple people detectors with different capabilities into a unified crowd analysis system. The first people detector provides rapid baseline counting, while the second people detector refines this data with enhanced accuracy for specific analysis periods. By combining these detectors' strengths, the system achieves high productivity in real-time crowd monitoring without requiring any single detector to perform all functions at maximum capacity, thus reducing overall processing resource requirements.
Data Source
AI summary
Aspects of the invention provide a vision pipeline-based method of censusing a crowd that includes presenting content on an outdoor digital display or other content player (e.g., a loudspeaker) and capturing with a video camera or other image acquisition device frames or other time-wise succession of images of a scene in the field of view or otherwise in the vicinity of the display/player. First and second people detectors (both, for example, face detectors) are used to determine respective counts of persons in the scene. Estimated viewing statistics, generated with a detection pipeline that includes the first people detector and a tracker, include at least one of (i) a number of persons in the scene that viewed the content on the player, and (ii) for at least one of those persons, a duration during which he/she was in the scene and/or looking toward the player. Corrected viewing statistics, generated as a function of the estimated viewing statistics and a difference between the counts determined by each of the first and second people detectors, are used to select or alter content presented on the player.