Multi-Camera Target Tracking With Dynamic Re-Identification Thresholds
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing multi-camera tracking systems face significant performance drops in real-world conditions, leading to high rates of false positives and negatives, necessitating excessive human interaction to validate or invalidate re-identification proposals.
Innovation Solution
A method for real-time multi-camera tracking that adjusts the human threshold based on the number of reference tracklets, allowing automated assignment or manual validation of new tracklets, and includes a phase to enrich the reference list with past tracklets, reducing operator interaction and improving tracking performance.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If deep learning models and re-identification models are used for multi-camera tracking, then tracking capability is provided, but performance drops significantly on real data leading to high false positive and false negative rates
Solution Approach 1:
The system performs preliminary actions by enriching the reference list with past tracklets before tracking begins. This pre-processing step accumulates historical tracking data to create a more robust reference base, which compensates for the domain drift problem and improves subsequent tracking accuracy without requiring real-time human intervention.
Solution Approach 2:
The patent implements dynamic threshold adjustment where the human threshold is adapted based on the number of reference tracklets available. As more reference tracklets are accumulated, the threshold dynamically adjusts to optimize the balance between automated tracking and human validation, thereby improving reliability while maintaining measurement precision.
2Reliability
If a fixed human threshold is used for validating re-identification proposals, then validation criteria are maintained, but excessive human interaction is required to validate or invalidate proposals
Solution Approach 1:
The system replaces fixed threshold validation with dynamic threshold adjustment. The human threshold is automatically adapted based on the quantity of reference tracklets in the reference list, allowing the system to optimize validation criteria in real-time. This reduces unnecessary human interaction while maintaining validation reliability by adjusting strictness according to available reference data.
Solution Approach 2:
The system performs self-service by automatically managing the validation process through dynamic threshold adjustment. Instead of relying on operators to consistently apply fixed thresholds, the system autonomously adapts validation criteria based on accumulated reference data, reducing operator burden while maintaining consistent validation reliability.
3Speed
If the number of reference tracklets is small, then tracking initialization is faster, but the human threshold must be higher requiring more human intervention
Solution Approach 1:
The system performs preliminary enrichment of the reference list with past tracklets before tracking begins. This pre-accumulation of reference data allows the system to start with a more robust foundation, enabling better automated tracking performance without requiring excessive human intervention, thus resolving the trade-off between initialization speed and automation capability.
Solution Approach 2:
The dynamic human threshold adjustment mechanism allows the system to adapt its automation level based on the number of reference tracklets. When reference tracklets are scarce, the threshold adjusts to maintain appropriate human oversight, while automatically increasing automation capability as more reference data becomes available, thus balancing speed and automation extent dynamically.
Data Source
AI summary
The invention relates to a method for real-time multi-camera tracking of a target object comprising, for at least one new tracklet captured by at least one camera, calculating at least a first distance between the new tracklet and at least one reference tracklet belonging to the target object. The method also includes, when the smallest calculated first distance is between a first predetermined threshold value, called re-identification threshold, and a second predetermined threshold value, called human threshold, presenting the new tracklet to an operator, for the operator to assign or not assign the new tracklet to the target object, manually. The human threshold is a function of the number of reference tracklets in the reference list. The invention also relates to a computer program, to an object re-identification device and to an object re-identification system implementing such a method.