Satellite 3D Reconstruction With Geocoded ML Depth Mapping
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing satellite imagery has lower resolution and is time-consuming to capture large areas, making efficient and accurate 3D reconstruction challenging.
Innovation Solution
A method using deep learning with a trained Machine Learning Network (MLN) to predict depth maps from overlapping 2D satellite images, incorporating imaging device parameters and multi-view geocoded ground truth data for improved 3D reconstruction.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If satellite images are used to capture large areas, then coverage area and time efficiency are improved, but image resolution deteriorates
Solution Approach 1:
A trained machine learning network is introduced as an intermediary between the low-resolution satellite images and the final 3D reconstruction. The network processes the satellite images and outputs enhanced depth maps that contain detailed geometric information, effectively mediating between the limited resolution of satellite data and the high precision requirements of 3D reconstruction.
Solution Approach 2:
The patent transforms the input from raw satellite images to processed depth maps by changing the data representation parameters. The machine learning network learns to map image pixels to depth values, fundamentally changing how the data is represented and processed to achieve both efficiency and precision.
2Device complexity
If traditional methods are used for 3D reconstruction from satellite images, then processing simplicity is maintained, but reconstruction accuracy and efficiency deteriorate
Solution Approach 1:
The patent replaces traditional mechanical/algorithmic image processing systems with a machine learning-based system. Instead of using conventional photogrammetry or stereo vision algorithms, a deep learning network is employed to directly predict depth maps from satellite images, achieving superior accuracy and efficiency.
Solution Approach 2:
The machine learning network is trained on synthetic data that copies and augments the characteristics of real satellite imagery. This synthetic training data allows the network to learn robust patterns and relationships, enabling it to accurately reconstruct 3D scenes from the limited and noisy satellite image data.
Data Source
AI summary
The present disclosure generally relates to a computer-implemented method for use in 3D reconstruction from satellite imagery. The method includes training, by a computing device, a plurality of machine learning networks (MLNs) based on a training set including multi-view 3D geocoded ground truth data; inputting at least two partially overlapping 2D satellite images and imaging device parameters for the at least two partially overlapping 2D satellite images to each of the plurality of MLNs; for each of the plurality of MLNs, computing, by the computing device, a depth map of the at least two partially overlapping 2D satellite images based at least in part on the imaging device parameters; and generating at least one geocoded 3D surface model based, at least in part, on at least one of the depth maps.