Learning Data Generation Using Precise Map and Camera
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
The generation of large amounts of data for deep-learning applications, such as lane and road center line recognition, is time-consuming and costly, primarily due to the need for human annotation of ground truth data.
Innovation Solution
A device and method that utilize a camera, precise positioning sensor, and precise map database to automatically generate learning data by mapping shape information from the map to camera images, reducing the time and cost of data creation and improving learning performance.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If human annotation is used to generate ground truth data for deep-learning, then data accuracy is improved, but time consumption and cost increase significantly
Solution Approach 1:
The patent uses precise map data as a template or reference to automatically generate ground truth annotations for training images. Instead of manually annotating each image, the system copies geometric information from the precise map and applies it to corresponding images through automated matching algorithms, thereby eliminating manual annotation while maintaining high accuracy
Solution Approach 2:
The patent replaces the mechanical process of manual human annotation with an automated computational system. The system uses image processing algorithms, coordinate transformation, and automated feature matching to substitute the manual labor of human annotators, significantly reducing time consumption while maintaining annotation quality
2Manufacturing precision
If human annotation is used to generate ground truth data for deep-learning, then data quality is improved, but manufacturing cost increases
Solution Approach 1:
The system copies precise geometric information from pre-existing accurate maps and applies it automatically to training images. This copying approach eliminates the need for expensive manual annotation services while preserving high data quality, as the source map data is already professionally surveyed and verified
Solution Approach 2:
The system performs self-annotation by automatically extracting features from images and matching them with corresponding features from precise maps. The automated pipeline includes image preprocessing, feature detection, coordinate transformation, and annotation generation without human intervention, making the process cost-effective while maintaining quality through algorithmic precision
3Reliability
If large amounts of data are collected using a network for deep-learning, then learning performance is improved, but data generation complexity increases
Solution Approach 1:
The patent creates a multi-functional data generation system that can handle multiple tasks: image collection, precise positioning, map matching, automated annotation, and data formatting. This universal system replaces multiple separate processes (manual annotation tools, data collection networks, coordinate transformation tools) with a single integrated pipeline, reducing overall complexity while enabling large-scale data generation for improved learning performance
Data Source
AI summary
A vehicle may include a camera for capturing a region around the vehicle, a positioning sensor for measuring a position of the vehicle, a database for storing a precise map, and a learning data generating apparatus for generating data for learning based on the captured region, the position of the vehicle, and the precise map.