Surveillance System Person Re-Identification via Object Assignment
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Standard person re-identification techniques in video surveillance systems are error-prone due to variations in appearance across different cameras, making it difficult to track individuals accurately in large surveillance areas like retail environments.
Innovation Solution
A surveillance system that includes a camera network with a person detection module, an object detection module, and an action detection module, which assigns object information to individuals based on actions performed, allowing for enhanced re-identification by using object information across multiple surveillance subareas.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If standard person re-identification techniques are used, then the system can attempt to identify people, but the re-identification accuracy deteriorates due to appearance variations across different cameras
Solution Approach 1:
The patent introduces object information as an intermediary element to bridge the gap between camera views. Instead of directly comparing person appearances across cameras (which fails due to variations), the system uses objects as mediators - detecting objects in the surveillance area and assigning them to persons, then using this object assignment as the basis for re-identification. This intermediary approach resolves the contradiction by providing a stable reference frame that is not affected by camera-specific appearance variations.
Solution Approach 2:
The patent transitions from a single-dimension identification approach (appearance-based) to a multi-dimensional approach by adding object information as a new dimension. The system now identifies persons not only by their appearance but also by the objects they are assigned, creating a richer identification space that compensates for appearance variations across different cameras and surveillance areas.
2Ease of operation
If appearance-based re-identification is used, then the system can identify people, but it becomes error-prone when a person looks different in different cameras
Solution Approach 1:
The patent uses object information as an intermediary that is independent of camera-specific appearance variations. Objects detected in the surveillance area serve as stable references that can be reliably assigned to persons across different cameras, providing a reliable basis for re-identification that does not depend on consistent appearance recognition across different viewing conditions and camera perspectives.
Solution Approach 2:
The patent replaces the unreliable appearance-based recognition mechanism with an object-based assignment mechanism. Instead of relying on visual recognition of person appearances (which is sensitive to camera variations), the system uses object detection and assignment as a substitution mechanism that provides more stable and reliable re-identification results.
3Reliability
If only person appearance features are used for re-identification, then the system remains simple, but additional identification features are needed to improve accuracy
Solution Approach 1:
The patent merges multiple identification approaches by combining person detection with object detection and assignment. The system integrates appearance-based person identification with object-based identification, creating a hybrid approach that leverages the strengths of both methods. This merging provides additional identification features (object information) while maintaining the simplicity of the overall system architecture through unified processing.
Solution Approach 2:
The patent creates a multi-functional surveillance system where the same infrastructure (cameras, processing units) serves multiple purposes: detecting persons, detecting objects, assigning objects to persons, and performing re-identification. This universality allows the system to gain enhanced re-identification accuracy without proportionally increasing complexity, as existing components are made multi-functional rather than adding entirely separate systems.
Data Source
AI summary
A surveillance system having an interface to a camera network for video surveillance of a surveillance area. The camera network includes a plurality of cameras each for capturing a surveillance subarea. The cameras are designed to provide surveillance images of the surveillance subareas. The surveillance system also includes a surveillance device for re-identifying people in the surveillance images. The surveillance device includes a person detection module for detecting people and an object detection module 10 for detecting objects. The surveillance device 5 includes an assignment module 12 designed to assign at least one item of object information to a person 4, and an action detection module 11 designed to detect an action of the person 4 on one of the objects 8.