Crowd Congestion Analysis for Circumvention Detection
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing methods for analyzing behavior in images, particularly in crowded areas, are limited in accuracy when the number of people is small, and require specific camera angles and locations, making them difficult to apply in general settings like passages or roads.
Innovation Solution
A device and method that estimate the degree of crowd congestion in multiple partial areas of an image, using both distribution state and temporal transition of congestion to detect circumventing behavior, allowing for accurate detection regardless of the number of people and camera angle.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If movement information of pixels obtained by using optical flow is used to analyze crowd movements, then circumventing behavior can be detected in crowded areas (100 or more people), but the method cannot be applied when the number of people is small (less than twenty people) because the amount of movement information is reduced
Solution Approach 1:
The image is divided into multiple partial regions, and the degree of crowd congestion is estimated for each partial region independently. This segmentation allows the system to detect circumventing behavior in both crowded and non-crowded environments by analyzing local congestion patterns rather than relying on global crowd movement information.
Solution Approach 2:
The invention changes the parameter from movement information (optical flow) to congestion degree information. By estimating the degree of crowd congestion in each partial region and analyzing its distribution state and temporal transition, the system can detect circumventing behavior regardless of the total number of people in the scene.
2Quantity of substance
If a monitoring camera is determined to a considerably reduced angle of view to capture a crowd of 100 or more people, then circumventing behavior can be analyzed, but it becomes difficult to apply in general settings like passages or rooms where the number of people is small
Solution Approach 1:
By dividing the image into multiple partial regions and estimating congestion degree for each, the system can effectively analyze crowd situations in both wide-angle views (capturing many people) and narrow-angle views (capturing few people). The segmentation approach adapts to different camera angles and locations.
Solution Approach 2:
The invention creates a universal detection method that works across different environments (crowded areas, passages, rooms) and different camera configurations. The congestion degree estimation approach is applicable regardless of the number of people or camera angle, making it multi-functional.
3Measurement precision
If the camera angle is determined to capture a suspicious person in an enlarged manner, then the person can be identified, but it becomes hard to capture many people in a single image, making it difficult to apply crowd analysis methods
Solution Approach 1:
The image is divided into multiple partial regions, allowing the system to analyze congestion patterns even when only a few people are captured. Each partial region's congestion degree is estimated independently, enabling detection of circumventing behavior regardless of the total number of people in the frame.
Solution Approach 2:
The system changes from analyzing movement information (which requires many people) to analyzing congestion degree distribution and temporal transitions. This parameter change enables effective analysis even when the camera is zoomed in on a single person or small group.
Data Source
AI summary
A device (100) for detecting circumventing behavior includes an estimation unit (101) that estimates a degree of crowd congestion in relation to each of a plurality of partial areas of a target image, and a detection unit (102) that detects circumventing behavior of a crowd by using a distribution state and a temporal transition of the degree of congestion estimated by the estimation unit (101).