3D Semantic Image Generation for Realistic Geographic Viewpoints
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing technologies for generating images based on geographic information struggle with visualization at specific angles, lack of realism, and limited versatility, particularly in satellite images and three-dimensional models, and face challenges in utilizing reproduced data for training artificial neural networks.
Innovation Solution
A method and system that utilize an artificial neural network model to generate synthetic images by combining 3D semantic models with location and directional information, allowing for realistic terrain visualization from desired viewpoints, and enabling editing with various conditions, including weather, time, and camera movements.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Quantity of substance
If satellite image-based image generation technology is used, then terrain information is provided, but visualization at specific angles is difficult
Solution Approach 1:
The patent transforms two-dimensional satellite images into three-dimensional semantic models by adding depth and spatial information. This enables the system to generate images from multiple viewpoints and angles, resolving the contradiction between providing comprehensive terrain information and enabling visualization at specific angles.
Solution Approach 2:
The patent introduces a 3D semantic model as an intermediary between satellite images and final generated images. This intermediary structure preserves terrain information while enabling flexible viewpoint control, allowing the system to maintain both comprehensive terrain data and specific angle visualization capabilities.
2Adaptability or versatility
If three-dimensional model-based image generation technology is used, then specific viewpoint visualization is enabled, but realism is insufficient
Solution Approach 1:
The patent merges the advantages of satellite image-based terrain information with three-dimensional model-based viewpoint control. By combining 2D satellite data with 3D semantic models and using neural networks to generate realistic textures and lighting, the system achieves both specific viewpoint visualization and high realism.
Solution Approach 2:
The patent uses neural networks to dynamically adjust image parameters such as lighting, texture, and color based on the 3D semantic model and desired viewpoint. This parameter transformation enables the system to maintain realism while providing flexible viewpoint control that was previously mutually exclusive.
3Quantity of substance
If GIS is used, then detailed geographic data is obtained, but realistic representation is lacking
Solution Approach 1:
The patent replaces traditional GIS data processing with neural network-based image generation. Instead of directly displaying raw geographic data, the system uses deep learning models to transform detailed geographic data into realistic visual representations, maintaining data accuracy while improving aesthetic quality.
4Quantity of substance
If existing geographic information is used, then data collection is possible, but utilization for training neural networks is difficult
Solution Approach 1:
The patent transforms geographic data from its original format into a format suitable for neural network training by generating synthetic images with annotated semantic information. This parameter transformation creates training data that maintains geographic accuracy while being directly compatible with modern deep learning frameworks.
Data Source
AI summary
A method for generating a synthetic image includes obtaining first location information and first directional information associated with the first location information, obtaining a three-dimensional (3D) semantic model associated with the first location information and the first directional information, generating first content information representing structural information of objects to be generated in a first synthetic image based on the first location information, the first directional information, and the 3D semantic model, generating the first synthetic image based on the first content information using an artificial neural network model, and outputting the generated first synthetic image.