Subject Re-Identification Feature Vectors for Cross-Camera Tracking
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing systems struggle to accurately track and re-identify subjects in real space, particularly in crowded areas with high entry and exit frequencies, due to occlusions, temporary obscuration, and errors in tracking, leading to issues like swapped identifiers and missed subjects.
Innovation Solution
A system using multiple cameras with overlapping fields of view generates reidentification feature vectors through a trained machine learning model to match subjects across time intervals, correcting errors by calculating similarity scores and reassigning tracking identifiers.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If multiple cameras with overlapping fields of view are used to track subjects, then subject tracking coverage is improved, but system complexity increases
Solution Approach 1:
The patent combines data from multiple cameras with overlapping fields of view into a unified tracking system. The system merges subject identifiers and tracking data across different camera views, using image recognition engines to correlate subjects across camera boundaries and maintain consistent tracking throughout the monitored area.
Solution Approach 2:
The patent introduces subject reidentification engines as intermediary components that mediate between multiple camera tracking systems. These engines generate reidentification feature vectors and calculate similarity scores to determine whether subjects detected by different cameras are the same individual, enabling seamless cross-camera tracking.
2Measurement precision
If subject reidentification using machine learning models is implemented, then tracking accuracy is improved, but computational processing time increases
Solution Approach 1:
The patent pre-generates reidentification feature vectors for detected subjects and stores them for quick retrieval. When a subject needs to be reidentified across cameras, the system retrieves pre-computed feature vectors and calculates similarity scores, avoiding the need to re-run complex machine learning models in real-time.
Solution Approach 2:
The patent extracts only the essential reidentification features from full subject images and processes only these extracted features for matching. The system takes out key discriminative features from complete images, reducing the computational burden while maintaining identification accuracy.
3Duration of action of stationary object
If continuous subject tracking is maintained across all areas, then subject persistence is improved, but system resource consumption increases
Solution Approach 1:
The patent implements dynamic tracking where the system adjusts tracking intensity based on subject movement and camera coverage. Subjects are continuously tracked only in areas where they are detected, and the system dynamically updates subject identifiers and positions based on available camera data, reducing resource consumption in areas with no active subjects.
Data Source
AI summary
Systems and methods for re-identifying tracked subjects in an area of real space are disclosed. The method includes generating first and second reidentification feature vectors of a first subject identified from a first time interval from the respective first and second sequences of images as obtained from the first time interval, and generating third and fourth reidentification feature vectors of a second subject identified from a second time interval from the respective first and second sequences of images. The method includes calculating a first similarity score between the first and third reidentification feature vectors, and calculating a second similarity score between the second and fourth reidentification feature vectors. The method includes matching the second subject to the first subject when at least one similarity score is above a pre-defined threshold.