Multi-Object Tracking via Integrated Similarity Neural Network
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current multi-object tracking technologies face challenges with irregular motion, incorrect detection, and obstruction due to obstacles, leading to degraded tracking accuracy when using convolutional neural networks for similarity evaluation in tracking-by-detection approaches.
Innovation Solution
An integrated similarity neural network is employed, utilizing a Siamese neural network structure with convolutional neural networks to evaluate similarity between target objects and candidates, and applying the Hungarian algorithm to allocate candidates based on similarity values, enabling robust tracking across multiple objects.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If convolutional neural network features are used for similarity evaluation in tracking-by-detection approach, then accuracy of similarity evaluation between objects is improved, but tracking accuracy is degraded due to irregular motion, incorrect detection results, and obstruction
Solution Approach 1:
The patent merges single object tracking algorithm and multi-object tracking algorithm into a unified tracking system. The single object tracking algorithm handles cases with irregular motion and obstruction by searching the perimeter of objects, while the multi-object tracking algorithm handles distinction between multiple objects. This combination resolves the contradiction by maintaining high similarity evaluation accuracy while improving overall tracking reliability through complementary algorithms.
Solution Approach 2:
The tracking system is designed to perform multiple functions: it can operate as a single object tracker when objects are obscured or moving irregularly, and as a multi-object tracker when multiple objects need to be distinguished. This multi-functional approach allows the system to maintain reliable tracking accuracy across diverse scenarios while preserving the high accuracy similarity evaluation capability of convolutional neural networks.
2Measurement precision
If single object tracking algorithm is directly applied to multi-object tracking, then object distinction capability is improved, but identification switching occurs frequently and separate training is required
Solution Approach 1:
The patent combines single object tracking and multi-object tracking algorithms into a unified framework where they work together rather than separately. The system dynamically switches between or combines their operations based on scene requirements, eliminating the need for separate training processes while maintaining the object distinction capability of multi-object tracking and the robustness of single object tracking.
Data Source
AI summary
Disclosed is a method of tracking multiple objects included in an image using a multi-object tracking apparatus including an integrated similarity neural network, the method including setting a tracking area in an input image, extracting at least one object candidate for a target object from the tracking area; extracting reference features for the target object, the object candidate, and the tracking area, selecting two of the target object, the object candidate, and the tracking area to evaluate similarity based on the reference features; allocating the object candidate to the target object on the basis of the evaluated similarity; and tracking the target object on the basis of a location of the allocated object candidate.