Subspace Rotation for Real-Time Image Change Detection
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
The singular spectrum transformation method for image change detection has a high calculation cost due to singular value decomposition, which hinders real-time detection of image changes.
Innovation Solution
An image change detection device that calculates the degree of change between subspaces and updates the first subspace to approach the second subspace when no change is determined, allowing for subsequent change detection without performing singular value decomposition on the next comparison period's image.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If singular value decomposition is performed on images in periods for comparison to generate subspaces for change detection, then change detection accuracy is improved, but calculation time increases significantly
Solution Approach 1:
The patent performs singular value decomposition and subspace generation in advance for images in periods for comparison, storing the results beforehand. This preliminary action allows that the computational heavy lifting is done before the actual change detection is needed, so when change detection is required, only lighter operations are needed to compare the pre-computed subspaces.
Solution Approach 2:
The patent dynamically adjusts the approach based on whether change is detected. When no change is detected, the system uses the pre-computed subspaces directly without reperforming SVD. When change is detected, the system can then perform more comprehensive analysis. This dynamic adaptation allows the system to optimize calculation time based on the specific detection scenario.
2Reliability
If singular value decomposition is performed on all images in change detection periods, then detection reliability is improved, but processing speed decreases
Solution Approach 1:
The system performs SVD and subspace computation in advance for comparison period images, storing the results for later use. This preliminary computation ensures that when change detection is needed, the heavy computational work is already done, allowing faster and more reliable detection without repeating the expensive SVD operation on all images.
Solution Approach 2:
The patent extracts and stores only the essential subspace information (left singular vectors and singular values) from the full image data. By taking out only the necessary components for change detection rather than processing the complete images through SVD repeatedly, the system maintains detection reliability while significantly improving processing speed.
3Productivity
If the system performs change detection for all periods, then detection completeness is improved, but computational cost increases
Solution Approach 1:
The system performs SVD and subspace generation as a preliminary action for images in periods for comparison, storing these pre-computed subspaces. This allows the system to maintain detection completeness by being able to compare any period against the pre-computed reference subspaces without having to perform expensive SVD operations on all images during the actual detection process.
Solution Approach 2:
The patent creates a copy of the subspace information from comparison period images and stores it for later use in change detection. By copying the essential subspace data rather than repeatedly computing it through SVD, the system can perform comprehensive detection across all periods while significantly reducing the total computational cost.
Data Source
AI summary
This image change detection device includes a subspace updating unit that, when a determination unit determines that there has been no change, updates a first subspace by rotating the first subspace so that the first subspace approaches a second subspace. When the determination unit determines that there has been no change, a degree-of-change calculating unit calculates the degree of change between the first subspace updated by the subspace updating unit and a second subspace whose bases are left singular vectors of an image in a change detection period next to a change detection period.