Parametric Building Model Extraction from Georeferenced Images
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current methods for monoscopic extraction of buildings from aerial or spatial images, especially SAR images, are inefficient due to reliance on low-level primitives, sensitivity to image noise and radiometric characteristics, and require complex threshold settings, leading to errors and increased costs.
Innovation Solution
A method that uses a hypothesis-refutation approach starting from high-level primitives, employing optimization techniques to jointly determine all descriptive parameters of buildings, reducing errors and eliminating the need for multiple threshold settings by iteratively adjusting parameters based on radiometric homogeneity and contour adequacy criteria.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Manufacturing precision
If manual extraction methods are used to obtain building models from aerial or spatial images, then good quality models can be obtained, but the process becomes tedious and expensive
Solution Approach 1:
The system enables automatic building extraction by having the computational algorithm perform the extraction task autonomously without human intervention. The method uses optimization techniques to automatically determine building parameters from image data, replacing manual operator work with self-executing computational processes that maintain high accuracy while dramatically reducing time and cost.
2Productivity
If automatic extraction techniques based on low-level primitives are used, then extraction speed is improved, but errors increase due to sensitivity to image noise and radiometric characteristics
Solution Approach 1:
Instead of building up building models from low-level image primitives (edges, lines, regions), the invention inverts the approach by starting with high-level parametric building models and adjusting their parameters to match the image data. This top-down approach uses optimization to directly determine building parameters (position, dimensions, orientation) without being sensitive to noise in intermediate processing stages, thereby improving both speed and accuracy.
3Adaptability or versatility
If segmentation-based algorithms with multiple thresholds are used for building extraction, then extraction capability is enhanced, but device complexity and parameter tuning requirements increase
Solution Approach 1:
The invention changes the fundamental parameters being optimized from multiple segmentation thresholds to a small set of direct building model parameters (position coordinates, dimensions, orientation angles). By formulating the extraction as an optimization problem that directly adjusts building parameters to match image observations, the system maintains versatility in extracting buildings of various types while reducing complexity from managing multiple interdependent thresholds to optimizing a concise parameter set.
Data Source
Figure 1
Figure 2
Figure 3
AI summary
The present invention relates to a method for modeling real objects represented in an image of the Earth's surface, particularly buildings, from a geographically referenced image. The image is produced by an aerial or space-based sensor associated with a physical image capture model. The method comprises at least the following steps: ■ selecting a parametric model of the external surface of said real object (101); ■ for several sets of parameters of said model: o projecting (103) the parameterized model onto the image by applying the physical image capture model; o evaluating the fit (104) between the projected model and the radiometric characteristics of the image; ■ determining the model parameters for which the fit is best (106) for modeling said object with these parameters. The invention is particularly applicable to remote sensing, digital geography, and the creation or updating of 3D urban databases.