Crowd Head Localization Using DIoU Matching and Feature Pyramids
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing artificial intelligence models struggle to accurately localize heads of people in crowded images captured by CCTV cameras, leading to inadequate response to accidents due to significant errors in predicting crowd density and determining person positions.
Innovation Solution
A method involving training an AI model with non-replacement matching based on DIoU loss, label assignment, and feature pyramid structure to accurately detect head center coordinates using anchor points and ground truth points, and a computer program to execute this method.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If existing artificial intelligence models are used for crowd density prediction, then the system can provide crowd density estimates, but the measurement precision of head localization and person position determination is insufficient due to significant errors
Solution Approach 1:
The patent segments the head localization task into multiple processing stages: feature extraction from crowd images, anchor point generation at potential head locations, feature pyramid construction for multi-scale detection, and coordinate refinement through regression modeling. This segmented approach allows each stage to optimize for its specific function, improving overall measurement precision.
Solution Approach 2:
The patent introduces a feature pyramid structure that adds a dimensional aspect to head detection by creating multiple scaled versions of feature maps. This allows the model to detect heads at various sizes and distances simultaneously, significantly improving localization accuracy in crowded scenes where heads vary greatly in apparent size.
2Area of stationary object
If more CCTVs are installed to improve safety monitoring coverage, then the surveillance area is expanded, but the device complexity and operational burden increase due to lack of personnel to control them
Solution Approach 1:
The system implements self-service through automated AI processing. The crowd density prediction model and head localization algorithms automatically analyze CCTV footage without human intervention, extracting meaningful information (crowd density, head positions) directly from video streams. This allows extensive CCTV coverage to be managed with minimal personnel.
Solution Approach 2:
The AI system performs multiple functions simultaneously: it predicts crowd density, localizes individual heads, determines person positions, and can trigger alerts. This multi-functionality allows a single control system to manage extensive CCTV networks efficiently, reducing the operational burden despite increased coverage area.
3Manufacturing precision
If traditional crowd density prediction methods are used, then the processing speed is maintained, but the manufacturing precision of head localization is insufficient leading to inadequate response to accidents
Solution Approach 1:
The system performs preliminary action by continuously pre-processing and analyzing crowd images in real-time, maintaining updated crowd density predictions and head localization data before incidents occur. This allows the system to immediately provide accurate localization information when accidents happen, reducing response time without sacrificing precision.
Data Source
AI summary
The present invention proposes a method of localizing heads of people in a crowd, which is capable of localizing heads of people in a crowd appearing in an image captured by a camera with high accuracy. The method may include performing label assignment to train the AI model. The matching is performed in ascending order of a difference in probability of the head being present at a predicted point predicted from the AI model based on a distance IoU loss value between the anchor point and the ground truth points, and the anchor point. The present invention was carried out with the support of the Civil-Military Technology Cooperation Project conducted by the Civil-Military Cooperation Promotion Agency with funds from the government of the Republic of Korea (Ministry of Trade, Industry and Energy and Defense Acquisition Program Administration) (Project No. 23-CM-Al-15).