Neural Network Training Data Generation Using Map Data
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current methods for training neural networks to recognize roads from images require manual extraction of information, which is time-consuming and inefficient, and lack automated systems for generating accurate training data.
Innovation Solution
An automated method and apparatus for generating training data by acquiring vehicle images and location information, retrieving map data, determining truth data associated with the road, and using this data to train neural networks, which includes determining probabilities of road types, drivability, and gradients, and controlling vehicles based on recognized road information.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If manual extraction of information is used to generate training data, then accuracy of training data can be ensured, but time consumption and inefficiency increase
Solution Approach 1:
The patent uses map data as a reference copy to automatically generate truth data for training. Instead of manual extraction, the system copies road information from pre-existing accurate map databases and uses it to create ground truth labels for training images, thereby eliminating time-consuming manual work while maintaining accuracy through the reliability of map data
Solution Approach 2:
The patent replaces the mechanical manual extraction process with an automated computer-based system. The processor automatically retrieves map data, processes images, and generates training data without human intervention, substituting the manual mechanical process with an automated computational system that is both faster and equally accurate
2Productivity
If automated methods are used to generate training data, then productivity increases, but measurement precision may deteriorate
Solution Approach 1:
The automated system copies verified road information from authoritative map databases to generate truth data. This copying approach ensures that the automated process inherits the accuracy of established map data while achieving high productivity through automation
Solution Approach 2:
The patent introduces map data as an intermediary between the image processing system and the training data generation. The map data serves as a mediating reference that guides the automated extraction process, ensuring accuracy by comparing image features against known map information while maintaining high processing speed
3Reliability
If comprehensive map data processing is performed, then reliability of road information improves, but device complexity increases
Solution Approach 1:
The patent extracts only the necessary road information from comprehensive map data for training purposes. Instead of processing all available map data, the system selectively extracts relevant features such as road boundaries, intersections, and traffic signs that are essential for training the neural network, thereby reducing system complexity while maintaining reliability
Solution Approach 2:
The patent segments the map data processing into distinct components: retrieving map data, extracting road features, matching features with images, and generating training labels. This segmentation allows each component to be processed independently and efficiently, reducing overall system complexity while ensuring comprehensive and reliable road information is captured
Data Source
AI summary
A training data generation method includes acquiring an image captured from a vehicle and location information of the vehicle corresponding to the image; acquiring map data corresponding to the acquired location information; determining truth data including information associated with a road included in the image from the acquired map data; and generating training data including the image and the determined truth data.