People Counting via Foreground Baseline Head Detection
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing people counting techniques face challenges in high-traffic environments due to inaccurate foreground separation, occlusion, and perspective distortion, leading to errors and slowed processing speeds, especially in complex backgrounds.
Innovation Solution
A method and apparatus that detect circular contours as candidate heads by a foreground contour crossing a baseline, track these candidates through a search area, and count them for accurate head detection, minimizing errors and processing time.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If conventional foreground separation and object tracking is used, then the system works acceptably with low traffic or simple background, but the measurement precision deteriorates in complicated environment with high-traffic or dense population
Solution Approach 1:
The patent segments the image processing task by detecting foreground contours first, then identifying circular contours within those foreground regions. This two-stage segmentation approach isolates potential head locations to only those areas where people are actually present, eliminating false detections from background elements and improving measurement precision in high-traffic environments
Solution Approach 2:
The patent applies local quality by performing circular contour detection exclusively within foreground contour regions rather than across the entire image. This localized approach concentrates computational resources on areas containing people, enhancing head detection precision while maintaining reliability in complicated environments with dense populations
2Quantity of substance
If conventional object tracking is performed on all detected objects, then the system can track multiple objects, but the processing speed slows down significantly with many people passing through
Solution Approach 1:
The patent extracts only the circular contours from the foreground regions as candidate heads, discarding other foreground objects. This extraction principle reduces the number of objects requiring tracking from all foreground detections to only potential heads, thereby maintaining the ability to track multiple people while significantly improving processing speed in high-traffic scenarios
3Ease of operation
If foreground separation is performed to identify people, then objects can be tracked, but counting becomes error-prone due to shadow, occlusion or the like
Solution Approach 1:
The patent uses circular contour detection to identify heads, exploiting the characteristic spherical shape of human heads. This geometric approach is robust to shadows and occlusions because it detects the fundamental circular shape rather than relying on detailed texture or color information, thereby maintaining counting accuracy while preserving object tracking capability
4Quantity of substance
If regression is used to estimate number of objects, then counting can be performed, but the camera lens generates perspective distortion that wildly fluctuates depending on the camera and surrounding environment
Solution Approach 1:
The patent replaces regression-based estimation with direct circular contour detection and counting. This substitution eliminates dependence on perspective geometry and camera parameters, making the system adaptable to different cameras and environments without requiring recalibration or complex distortion correction models
Data Source
AI summary
A method for counting a number of people includes detecting a foreground contour from a current frame inputted, performing a head detection by defining a baseline virtually at a point perpendicular to a center point of a camera-captured area, by defining a search area around the baseline, and by detecting, as as candidate heads, circular contours to be recognized as heads of people at a point where the foreground contour crosses the baseline, tracking the candidate heads respectively in subsequent frames until the candidate heads pass through the search area, and counting up the number of people by the candidate heads that respectively pass through the search area.