Map Building Using Machine Learning and Image Processing
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing map creation methods are time-consuming, costly, and prone to errors due to manual data collection and updating, leading to degraded accuracy, especially when dealing with frequently changed buildings and roads.
Innovation Solution
An apparatus using machine learning and image processing, specifically deep learning based on convolutional neural networks, automatically builds maps by processing vector and raster data, generating ground-truth images, and predicting polygons to improve accuracy and reduce human error.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If manual data collection and map creation methods are used, then users can personally create and update maps, but it takes a long time to build maps, production costs are high, and accuracy is degraded due to user errors
Solution Approach 1:
The patent replaces the manual mechanical process of map creation with an automated image processing system using convolutional neural networks. The system automatically extracts building and road information from satellite images through deep learning, eliminating the need for manual data collection and processing, thereby significantly reducing time and improving accuracy
Solution Approach 2:
The system enables self-service map creation by automatically processing satellite images and generating map data without human intervention. The convolutional neural network performs autonomous feature extraction and classification, allowing the system to update maps independently based on newly acquired satellite imagery
2Ease of manufacture
If manual data collection methods are used, then maps can be created with human oversight, but production costs are high and updates are difficult when buildings and roads frequently change
Solution Approach 1:
The patent replaces complex manual data processing operations with an automated deep learning system. The convolutional neural network handles the complexity of interpreting satellite images, extracting features, and updating map data automatically, making the update process simple and efficient despite the underlying computational complexity
3Productivity
If automated machine learning methods are used, then map building time is reduced and accuracy is improved, but the system complexity increases
Solution Approach 1:
The patent implements a universal deep learning system that performs multiple functions: satellite image processing, building extraction, road extraction, and map data generation. The convolutional neural network serves as a multi-functional platform that handles various map creation tasks, reducing the need for separate specialized systems and making the complexity manageable through consolidation
Data Source
AI summary
An apparatus for building a map is disclosed. The apparatus for building a map according to the present invention includes a data collection unit configured to separately collect vector data and raster data, a vector data processing unit configured to generate ground-truth images of a previously set size by processing the vector data, a raster data processing unit configured to generate divided raster images of the set size by processing the raster data, and a polygon generation unit configured to generate predicted polygons through machine learning of the ground-truth images and the divided raster images and generate polygons which can be applied to map building based on the raster data on the basis of the predicted polygons and the ground-truth images.