Credit Allocation Network for Single-Target Tracking
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing target tracking methods face challenges in maintaining adaptability and robustness due to complex background disturbances, leading to contamination of the target appearance model by unreliable memory samples.
Innovation Solution
A single-target tracking method based on a credit allocation network is introduced, which evaluates the tracking state through a credit allocation network to prevent contamination of the target appearance model and improves memory sample quality using a new memory selection strategy.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If memory samples are updated indiscriminately during tracking, then the tracking model can adapt to target changes, but the target appearance model becomes contaminated by low-quality samples
Solution Approach 1:
The patent implements a feedback mechanism through the credit allocation network that evaluates the quality of memory samples before updating the appearance model. The credit score serves as a feedback signal to determine whether to update or discard samples, preventing contamination while maintaining adaptability
Solution Approach 2:
The credit allocation network acts as an intermediary between memory samples and the appearance model. It evaluates samples and assigns credit scores, serving as a gatekeeper that filters low-quality samples before they can contaminate the appearance model, thus resolving the contradiction between adaptability and reliability
2Adaptability or versatility
If the target template is always updated from the previous frame, then the tracking model adapts quickly to target changes, but it fails when the target is deformed, blocked, or out of sight
Solution Approach 1:
The patent prepares multiple historical memory samples in advance that represent different states of the target (including occluded and deformed states). When the current template fails due to occlusion or deformation, the system can retrieve and use pre-prepared alternative samples from memory, maintaining reliability while preserving adaptability through the credit-based selection mechanism
3Loss of information
If historical frames are used to obtain target information, then the tracking model gains more target information, but low-quality samples contaminate the appearance model
Solution Approach 1:
The credit allocation network provides feedback on the quality of each historical sample before it is used to update the appearance model. This feedback mechanism ensures that only high-quality samples (those with high credit scores) are incorporated, preventing information loss while maintaining model purity
Solution Approach 2:
The patent changes the parameter of sample selection from indiscriminate updating to credit-score-based selective updating. By introducing the credit score parameter, the system can differentiate between high-quality and low-quality samples, allowing effective use of historical information while preventing contamination
Data Source
AI summary
The present disclosure discloses a single-target tracking method based on a credit allocation network. The credit allocation network is at first provided, which can be updated online using a guiding focus loss function. The credit allocation network generates credit scores for the tracking results by learning features of the target object, which ensures that the reliable samples are updated for storage in the memory pool. In order to better adapt to the changes in the appearance of the target during the tracking process, a new memory selection strategy is provided to collect high-quality tracking results as the memory samples during the tracking process, which further improves the reliability and adaptability of the memory pool. The present disclosure uses the guiding focus loss function to update the credit allocation network online, which can select more reliable memory samples for the memory pool and improve the robustness of the tracking results.