Image Tracking Weight Maps for Temporal Appearance Changes
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing tracking methods using feature correlation, such as Li's Siamese Region Proposal Network, struggle to maintain accuracy when the subject undergoes time-series changes in an image.
Innovation Solution
An image processing apparatus that generates a weight map for feature correlation based on time-series images, using a weight map generation model to emphasize regions with minimal change, improving tracking robustness and accuracy by weighting features differently across image positions.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If feature correlation methods are used for tracking, then tracking speed is maintained, but tracking accuracy deteriorates when the subject undergoes time-series changes
Solution Approach 1:
The patent applies local quality by generating a weight map that assigns different weights to different spatial positions of the feature map. Regions with minimal temporal change receive higher weights, while regions with significant changes receive lower weights. This spatially varying weighting strategy allows the tracker to focus on stable, discriminative regions while suppressing regions that undergo temporal transformations, thereby maintaining tracking accuracy under appearance changes.
2Reliability
If uniform feature weighting is applied, then computational simplicity is maintained, but tracking robustness deteriorates when appearance changes occur
Solution Approach 1:
The patent implements preliminary action by pre-training a weight map generation model using supervised learning with ground truth weight maps. This pre-trained model automatically learns to generate appropriate weight maps that highlight stable regions and suppress changing regions. During actual tracking, the pre-trained model quickly generates weight maps without requiring complex real-time optimization, thus achieving robust tracking with moderate computational complexity.
Data Source
AI summary
An image processing apparatus is provided. The image processing apparatus searches for a tracking target from a target image. A feature of the target image and a feature of a first image of the tracking target at a first time are acquired. A weight of a feature for each of a plurality of positions of the first image of the tracking target is generated on the basis of the first image of the tracking target and a second image of the tracking target at a second time. The tracking target is detected from the target image on the basis of a correlation between a feature of the target image and a feature of a first image of the tracking target weighted on the basis of the weight.