Tracking Model Switching for Occlusion-Robust Image Tracking
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing tracking technologies struggle with maintaining accuracy when a tracking destination is partially masked by another object, leading to potential misidentification with similar objects and disrupted tracking.
Innovation Solution
An image processing apparatus that switches between tracking models based on the detection of masking and its termination, using a first model for the masking object and a second model for the original tracking target when masking occurs and ends, respectively.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If a tracking model lowers a threshold value to track a masking material as a temporary tracking target, then tracking can continue when the original tracking destination is masked, but the temporary tracking target may incorrectly transition to another object with similar appearance
Solution Approach 1:
The patent applies dynamics by switching between different tracking models (first tracking model and second tracking model) based on the masking state. When masking is detected, the system dynamically transitions from the original tracking model to a first tracking model optimized for masking materials, and when masking ends, switches back to the original model. This dynamic adaptation resolves the contradiction by adjusting the tracking approach according to real-time conditions, maintaining both continuity and accuracy.
Solution Approach 2:
The patent changes parameters by using different threshold values and tracking criteria for different tracking models. The first tracking model uses a lowered threshold value suitable for detecting masking materials, while the original tracking model uses a higher threshold for accurate target identification. By changing these parameters based on the masking state, the system maintains tracking continuity without sacrificing accuracy when the original target is visible.
2Reliability
If multiple feature points are set for the tracking destination to enable tracking when partially masked, then tracking continuity is improved, but the system becomes more complex and may still fail when similar objects appear
Solution Approach 1:
The patent applies segmentation by dividing the tracking task into distinct segments: using the original tracking model for normal tracking, switching to the first tracking model when masking is detected, and returning to the original model when masking ends. This segmentation approach simplifies the overall system compared to using multiple feature points simultaneously, while maintaining tracking continuity through model switching rather than complex multi-point tracking.
3Reliability
If prediction using movement vector is used to continue tracking when masked, then tracking continuity is improved, but accuracy decreases when the masking object has similar appearance to the target
Solution Approach 1:
The patent introduces an intermediary mechanism - the first tracking model - that acts as a mediator between the original tracking model and the masking situation. When masking is detected, the first tracking model takes over to track the masking material, preventing incorrect predictions that would occur with movement vector methods. This intermediary approach maintains continuity while avoiding the accuracy problems of prediction-based methods.
Data Source
AI summary
An image processing apparatus comprises a tracking unit configured to perform a tracking process, using a tracking model, in which a tracking target in a captured image is tracked, and a switching unit configured to switch the tracking model to a first model that tracks a second object as the tracking target when masking of the first object by the second object is detected while the tracking unit tracks the first object as the tracking target, and to switch the tracking model to a second model that tracks the first object as the tracking target when termination of the masking is detected.