Terrain Classification via Voting and Elevation Data
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing methods for classifying terrain types, such as water, face challenges in determining valid thresholds over larger areas, often misclassifying land as water due to shadowed areas and requiring stereo matching that restricts image usage and computational efficiency.
Innovation Solution
A method that calculates terrain type indices from multiple aerial images taken at different times and angles, using a voting mechanism and surface elevation data to reduce the impact of shadows and moving objects, without the need for stereo matching, allowing for more reliable and accurate classification of large areas.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If a single threshold is used for terrain classification, then the classification process is simple and fast, but the classification accuracy deteriorates over larger areas due to shadows and varying conditions
Solution Approach 1:
The patent divides the classification process into multiple independent stages: acquiring multiple images of the same area taken at different times, calculating terrain indices for each image separately, and then combining results through a voting mechanism. This segmentation allows each image to be processed independently with simple thresholds while achieving high overall accuracy through aggregation.
Solution Approach 2:
The patent employs periodic action by acquiring multiple images of the same area at different times (periodically). Each image provides an independent classification opportunity, and the voting mechanism aggregates these periodic measurements to overcome temporary conditions like shadows that affect single-image classification.
2Reliability
If stereo matching is used to improve classification accuracy, then the reliability of terrain type identification improves, but the device complexity and computational effort increase significantly
Solution Approach 1:
The patent extracts and removes the complex stereo matching step from the classification process. Instead of requiring complex 3D reconstruction and matching algorithms, it simply acquires multiple 2D images at different times and combines them through voting. This extraction of the essential function (multiple observations) without the complex mechanism (stereo matching) maintains reliability while reducing complexity.
Solution Approach 2:
The patent uses multiple inexpensive, easily acquired images taken at different times rather than complex stereo image pairs. Each image is a simple 2D capture that can be obtained from standard satellite or aerial sources, replacing the need for specialized stereo photography equipment and complex processing systems.
3Measurement precision
If images taken at the same time of year are used for stereo matching, then matching accuracy improves, but the adaptability of the method deteriorates due to seasonal changes in vegetation and snow
Solution Approach 1:
The patent deliberately uses images taken at different times (periodically) rather than requiring simultaneous images. This periodic acquisition at different seasons allows the method to adapt to varying conditions while the voting mechanism handles the temporal differences. The method becomes more versatile as it can process images from any time period.
Solution Approach 2:
The patent creates a universal classification method that works across different seasons, terrains, and imaging conditions. By using multiple images taken at different times with voting aggregation, the method becomes multi-functional and adaptable to various environments without requiring specialized image pairs or timing constraints.
4Measurement precision
If multiple images are processed individually and combined, then the accuracy of terrain classification improves by reducing shadow impact, but the computational effort increases
Solution Approach 1:
The patent segments the computational task into independent image processing steps followed by simple voting aggregation. Each image is processed independently to generate terrain indices, then the results are combined through counting votes for each terrain type. This segmentation allows efficient parallel processing and avoids the need for complex iterative algorithms that would consume more energy.
Data Source
Figure 1~2
Figure 3
Figure 4~5
AI summary
The invention relates to a method (400) for classifying a terrain type in an area. The method comprises the step of obtaining (410) a plurality of overlapping aerial images of the area. The method also comprises calculating (430) at least one terrain type index for each part of each of the aerial images which lies in the area, where the at least one terrain type index represents the terrain type. The method also comprises the step of determining (440) at least one terrain type index for each part of the area based on the calculated at least one terrain type index for each part of each of the aerial images; and the step of classifying (450) the parts of the area for which at least one pre-determined conditions is met as containing the terrain type, wherein at least one of the at least one predetermined condition relates to a value of the determined at least one terrain type index. The invention also relates to a system (600), a computer program and a computer program product