Inflow Source Attribute Assignment for Crowd Anomaly Detection
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current tracking systems face challenges in automatically analyzing people flow and detecting abnormal movements across multiple cameras, especially in environments with restricted movement paths, where most trajectories have similar characteristics, making it difficult to identify changes or anomalies in crowd movement.
Innovation Solution
The system assigns an inflow source attribute to each trajectory in the camera view, using techniques for tracking across cameras to estimate and visualize the inflow source, allowing for the detection of changes and anomalies by distinguishing trajectories based on their inflow sources, even if they have the same direction, and combines this with within-camera trajectory analysis to trace the cause of changes.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Difficulty of detecting and measuring
If tracking across cameras is used to estimate inflow sources, then the ability to detect changes and anomalies in crowd movement is improved, but the computational complexity and system complexity increase
Solution Approach 1:
The system segments the surveillance area into multiple camera views and divides the crowd analysis task into: (1) within-camera trajectory tracking for movement visualization, and (2) across-camera inflow source estimation for anomaly detection. By segmenting the problem this way, each component can focus on specific aspects, reducing overall system complexity while improving detection capability.
Solution Approach 2:
The patent introduces an intermediary database that stores extracted image features and tracking information from multiple cameras. This database acts as a mediator between the camera systems and the analysis algorithms, enabling efficient retrieval and comparison of crowd movement data across cameras without requiring direct complex coordination between all camera systems.
2Reliability
If image matching techniques are applied to track persons across cameras, then re-identification capability is improved, but computational power requirements increase
Solution Approach 1:
The system performs preliminary extraction and storage of image features from multiple camera views in advance. By pre-processing and storing these features in a database, the system can quickly retrieve and match features during actual anomaly detection without performing computationally intensive image matching in real-time, thus reducing instantaneous computational power requirements.
Solution Approach 2:
Instead of applying heavy image matching algorithms to all person images across all cameras, the system focuses computational resources on specific local tasks: extracting features from salient areas, matching only when necessary for inflow source determination, and visualizing trajectories only for detected anomalies. This localized approach reduces overall computational burden while maintaining re-identification reliability.
3Measurement precision
If manual selection of tracking targets is used, then tracking accuracy is improved, but automation level decreases
Solution Approach 1:
The system implements self-service automation by automatically detecting abnormal movements and selecting tracking targets without human intervention. The anomaly detection algorithm autonomously analyzes trajectory patterns, identifies deviations from normal crowd flow, and initiates detailed tracking and visualization of suspected anomalies, freeing the user from manual target selection while maintaining high tracking accuracy through automated intelligent selection.
Data Source
AI summary
Example implementations described herein involve an attribute of an inflow source that is assigned to each trajectory in the camera view, which is used for crowd analysis. The inflow source is estimated using techniques of tracking across cameras. If the inflow source is different even if the flow trajectories are the same direction, the trajectories are visualized in different styles. By using the attribute of the inflow source, it is possible to detect changes and anomalies in the crowd that were difficult to detect using the prior art.