Terrain Image Region Extraction for Map-Free Change Detection
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing methods for extracting changes in ground surface features from satellite and aerial images require map information corresponding to a specific time point, limiting the time interval for extraction and reducing accuracy due to infrequent map updates.
Innovation Solution
A region extraction apparatus and method that utilizes ground surface images at two arbitrary time points, employing a candidate extraction unit, categorization unit, and region extraction unit to identify and extract regions where objects have changed, using characteristic calculation, categorization, and shape determination to enhance accuracy.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If map information is used for change extraction, then extraction accuracy is improved, but the time interval for extraction is limited due to infrequent map updates
Solution Approach 1:
The patent creates a synthetic map by copying and processing aerial images taken at a first time point through categorization to generate land cover information. This synthetic map serves as a reference without requiring actual updated map information, enabling comparison with images from the second time point while overcoming the limitation of infrequent map updates.
Solution Approach 2:
The system performs preliminary categorization and land cover identification on the first aerial image to create the synthetic map before the change detection process. This preliminary action prepares the reference data structure in advance, allowing flexible time interval selection for the second image without being constrained by actual map update schedules.
2Speed
If simple image comparison is used, then processing speed is improved, but extraction accuracy deteriorates due to reflection intensity and shadow differences
Solution Approach 1:
The patent transforms images from the spatial domain to the frequency domain using Fourier transform, changing the representation parameters of the images. This parameter transformation enables comparison based on frequency characteristics rather than raw pixel values, effectively reducing the impact of reflection intensity and shadow variations while maintaining processing efficiency.
Solution Approach 2:
The system replaces direct pixel-by-pixel mechanical comparison with a frequency-domain analysis approach using Fourier transform. This substitution allows for more robust change detection by comparing spectral characteristics rather than raw intensity values, improving accuracy without significantly increasing processing complexity.
3Measurement precision
If categorization processing is added to image comparison, then extraction accuracy is improved, but device complexity increases
Solution Approach 1:
The patent employs a categorization model that performs multiple functions: it identifies land cover types, generates the synthetic map, and provides the basis for change detection. This multi-functional approach consolidates several processing tasks into a single unified system, improving accuracy without proportionally increasing overall system complexity.
Solution Approach 2:
The categorization model acts as an intermediary that processes the first aerial image to create the synthetic map and land cover information. This intermediary structure separates the complex categorization task from the direct image comparison process, allowing each component to be optimized independently while working together to achieve high extraction accuracy.
Data Source
Figure 1~2
Figure 3~4
Figure 5
AI summary
A region detection device 5 comprises a candidate extraction unit 10, a classification unit 20, and a region extraction unit 30. The candidate extraction unit 10 extracts differing variation regions between a first terrain image 1-1 in which the ground surface was photographed from the sky at a first time point and a second terrain image 1-2 in which the ground surface was photographed at a second time point differing from the first time point. The classification unit 20 estimates, in accordance with the location of the first terrain image 1-1, a first category for a first terrain which was photographed, calculates a first classification image depicting the relationship between the location of the first terrain and the first category, estimates, in accordance with the location of the second terrain image, a second category for a second terrain which was photographed, and calculates a second classification image depicting the relationship between the location of the second terrain and the second category. The region extraction unit 30 extracts from the variation regions an extraction region in which the first category and the second category in the same location as the first category satisfy a predetermined condition.