Depthmap Generation via Cell Segmentation and Multi-View Stereo
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Aerial images of structures lack geometry and depth information, which is essential for three-dimensional modeling and geometry-based rendering applications.
Innovation Solution
A computer-implemented method and system that divide a geographic area into coarse and fine cells, using multi-view stereo processing to generate 3D points and mesh models from low- and high-resolution images, ultimately creating depthmaps for each high-resolution image.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If multi-view stereo processing is applied to all high-resolution images of the entire geographic area, then depthmap quality and measurement precision are improved, but processing time and computational resources increase excessively
Solution Approach 1:
The geographic area is divided into multiple cells, and each cell is processed independently using multi-view stereo processing. This segmentation allows the system to process only the subset of images relevant to each cell rather than processing all images at once, significantly reducing processing time while maintaining depthmap quality for each region.
Solution Approach 2:
The system performs preliminary processing by identifying which images are visible for each cell before executing multi-view stereo processing. This preliminary action of determining image-cell associations allows the system to pre-filter the input images, avoiding unnecessary processing of images that do not contribute to specific cells, thus reducing overall processing time.
2Measurement precision
If multi-view stereo processing is applied to all high-resolution images of the entire geographic area, then depthmap quality is improved, but memory requirements increase excessively
Solution Approach 1:
By dividing the geographic area into cells and processing each cell independently, the system loads and processes only the images relevant to each cell in memory at any given time. This segmentation prevents the system from requiring all images to be loaded into memory simultaneously, thus reducing peak memory requirements while maintaining the quality of depthmaps through proper multi-view stereo processing.
3Productivity
If the entire geographic area is processed as a single unit, then processing simplicity is maintained, but processing efficiency and productivity decrease
Solution Approach 1:
The system segments the geographic area into multiple cells that can be processed in parallel. This segmentation increases processing efficiency by enabling concurrent processing of multiple regions, thereby improving overall productivity. The complexity increase is managed through systematic organization of cells and their associated images, making the processing pipeline structured and manageable.
Solution Approach 2:
The system implements a universal processing framework that handles multiple cells using the same multi-view stereo processing algorithm. This multi-functional approach allows the same processing logic to be applied repeatedly across different cells, improving efficiency through reuse of processing routines while maintaining consistency in depthmap quality across the entire geographic area.
Data Source
AI summary
Systems and methods for generating a depthmap for images of structures in a geographic area are provided. In some aspects, a method includes dividing the area into cells. Each of the cells is visible in no more than a threshold number of images. The method also includes: generating, using multi-view stereo processing, a set of three-dimensional (3D) points of structures in each of the cells based on the images associated with a corresponding one of the cells; generating a mesh model of the structures in each of the cells based on the set of 3D points for a corresponding one of the cells; identifying one or more of the mesh models visible in a corresponding one of the images; and generating a depthmap for each of the images based on the identified one or more mesh models visible in a corresponding one of the images.