Map Geometry Generation Using Contrastive Data Conflation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional digital maps suffer from missing or erroneous lane geometry data, which can lead to inaccurate route guidance and reduced effectiveness of vehicle autonomy due to the reliance on manual verification and correction processes that are laborious and costly.
Innovation Solution
A method and apparatus that utilize data aggregation and conflation techniques, employing neural networks to process multiple data sources such as aerial images, LiDAR, and vehicle sensors, to generate and update map geometry by aligning embeddings using a contrastive loss function, thereby producing consistent and accurate map data.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If manual verification and correction processes are used to ensure map geometry accuracy, then map data reliability is improved, but labor cost and time consumption increase
Solution Approach 1:
The patent replaces manual verification and correction processes with an automated neural network-based system. The system uses encoders to process observation data from multiple sources, establishes contrastive loss between different data sources, and automatically generates corrected map geometry, eliminating the need for manual intervention while maintaining high accuracy
Solution Approach 2:
The map generation system performs self-verification and self-correction by using multiple data sources (aerial images, LiDAR, probe data) to cross-validate each other. The contrastive loss mechanism automatically identifies and corrects inconsistencies between different data sources without external manual input
2Manufacturing precision
If multiple data sources are integrated to improve map geometry accuracy, then manufacturing precision is improved, but device complexity increases
Solution Approach 1:
The patent employs a universal encoder architecture that can process multiple types of observation data (aerial images, LiDAR data, probe data) through the same neural network framework. This multi-functional approach allows diverse data sources to be integrated using a single processing pipeline, managing complexity while improving accuracy
Solution Approach 2:
The system transforms different types of observation data into a unified embedding space through neural network encoders. By changing the representation parameters of diverse data sources into consistent embedding vectors, the system enables accurate comparison and integration while maintaining manageable system complexity
3Productivity
If automated map generation from multiple data sources is implemented, then productivity is improved, but measurement precision requirements increase
Solution Approach 1:
The patent implements a feedback mechanism through contrastive loss calculation between embeddings from different data sources. The system continuously measures the consistency between multiple observation data sources and uses this feedback to adjust and align the embeddings, ensuring high measurement precision while maintaining automated high-speed processing
Solution Approach 2:
The neural network encoders act as intermediaries that transform raw observation data from multiple sources into a common embedding language. This intermediary representation layer enables precise comparison and alignment between different data types without requiring direct high-precision alignment of the original heterogeneous data
Data Source
AI summary
A method is provided for automatically creating road and lane geometry from images representing probe data within a geographical area using data aggregation and conflation. Methods may include: receiving first observation data associated with a geographic area; processing the first observation data through a first encoder to produce first embeddings; receiving second observation data associated with the geographic area; processing the second observation data through a second encoder to produce second embeddings; establishing contrastive loss between the first embeddings and the second embeddings; producing consistent embeddings from the first embeddings and the second embeddings using the contrastive loss; generating, from the consistent embeddings, map data reflecting the first observation data and the second observation data; and updating a map database with the map data.