Object Tracking Template Suppression for Shape-Aware Precision
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing object tracking systems face challenges in accurately tracking objects of varying shapes and sizes due to the imbalance between target object information and context information, leading to performance degradation, especially when the context information occupies a significant proportion in the template image.
Innovation Solution
The method involves setting a suppressed region in the template image based on the shape of the target box, refining the template feature map by suppressing the influence of context information, and updating the template based on predefined conditions to maintain tracking accuracy and stability.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If context information is included in the template image to improve tracking robustness, then tracking stability is improved, but tracking precision deteriorates due to information imbalance
Solution Approach 1:
The template image is segmented into multiple regions based on target box shape: when the target box is wider than it is tall, the upper and lower regions are suppressed; when the target box is taller than it is wide, the left and right regions are suppressed. This segmentation allows selective retention of relevant context information while eliminating regions that cause information imbalance and tracking precision degradation.
Solution Approach 2:
Different regions of the template image are assigned different qualities or weights. The suppression mechanism applies local adjustments to specific regions based on target shape characteristics, making the template representation more locally accurate by emphasizing regions containing relevant target information while de-emphasizing regions with excessive background context.
2Reliability
If the template image includes both target object information and context information, then tracking robustness is improved, but device complexity increases due to shape-based suppression logic
Solution Approach 1:
The suppression regions are dynamically adjusted based on the shape characteristics of the target box. The system automatically adapts which regions to suppress by comparing the width and height of the target box and applying appropriate suppression patterns, making the processing adaptive rather than static while maintaining reasonable complexity through simple geometric comparisons.
Data Source
Figure 1
Figure 2
Figure 3
AI summary
A processor-implemented object tracking method includes: setting a suppressed region in a template image based on a shape of a target box of the template image; refining a template feature map of the template image by suppressing an influence of feature data corresponding to the suppressed region in the template feature map; and tracking an object by determining a bounding box corresponding to the target box in a search image based on the refined template feature map.