Object Re-identification Using Relative Velocity Estimation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current re-identification tracking algorithms in video surveillance systems face inefficiencies due to reliance on pixel-based features, high computational resource usage, and high failure rates when tracking targets across non-overlapping camera views, especially due to image artifacts and large candidate object comparisons.
Innovation Solution
A method that estimates the relative velocity of target objects to reduce the number of candidate objects and images processed by determining the travel time and using object density measurements to refine candidate selection, thereby improving tracking efficiency and accuracy.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If pixel-based features are used for re-identification tracking, then the system can process images from multiple cameras, but the computational resource usage increases and tracking accuracy decreases due to image artifacts
Solution Approach 1:
The patent extracts only the essential feature - relative velocity - from the complex pixel-based feature set. By removing unnecessary pixel-level details that are prone to artifacts and high computational cost, the system retains only the velocity information that is critical for tracking targets across non-overlapping camera views, thereby reducing computational resources while maintaining or improving tracking accuracy
Solution Approach 2:
The patent changes the feature parameter from pixel-based spatial information to velocity-based temporal information. This parameter transformation allows the system to ignore static image artifacts and focus on dynamic motion characteristics, which are more reliable for re-identification tracking across different camera perspectives and lighting conditions
2Reliability
If all candidate objects are compared for re-identification, then the correct tracking rate may improve, but the computational time and resources increase significantly
Solution Approach 1:
The patent applies partial action by comparing only a subset of candidate objects that satisfy the relative velocity criterion, rather than exhaustively comparing all candidate objects. This selective comparison approach maintains sufficient tracking accuracy by focusing on likely candidates while dramatically reducing computational time and resources
Solution Approach 2:
The patent performs preliminary filtering of candidate objects based on relative velocity before conducting detailed re-identification comparison. This preliminary action eliminates obviously unrelated candidates in advance, so that the subsequent detailed comparison is performed only on a reduced set of promising candidates, optimizing the balance between tracking accuracy and computational efficiency
3Area of stationary object
If sparse camera network is used for re-identification tracking, then the coverage area increases, but the tracking reliability decreases due to lack of overlapping fields of view
Solution Approach 1:
The patent introduces a temporal dimension (velocity over time) to compensate for the lack of spatial overlap between camera fields of view. By utilizing relative velocity information that persists across time, the system can reliably track targets through the gaps between non-overlapping camera views, effectively extending coverage area without sacrificing tracking reliability
Data Source
AI summary
The present invention relates in particular to a method for re-identification of a target object in images obtained from several image sources, wherein each of the image sources obtains images representing an area associated with the corresponding image source. After having identified a target object in images obtained from one of a pair of image sources, a relative velocity of the target object in comparison with other objects previously identified in images obtained from the one of the pair of image sources is estimated. Then, based on the estimated relative velocity of the target object, a correspondence between the target object identified in images obtained from the one of the pair of image sources and a candidate object represented in images obtained from the other image source of the pair of image sources is established.