Intersection Structure Modeling Using Bayesian Link Probability
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current methods for generating roadway maps lack precision in representing lane-level and intersection information, requiring significant manual effort and being inadequate for advanced driver safety systems like self-driving vehicles.
Innovation Solution
A method that combines road lane information from databases or estimation algorithms with vehicle trajectory data from sources like LIDAR, radar, and video to determine lane node locations, assess link probabilities using Bayesian Model Averaging, and filter links to create an accurate model of the intersection structure.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If manual annotation or test vehicle methods are used to generate maps, then map accuracy and detail can be improved, but significant amounts of manual work and preparation effort are required
Solution Approach 1:
The system uses existing GPS track data from normal vehicle operations to automatically generate intersection maps. The map generation process serves itself by utilizing data that vehicles already collect during regular use, eliminating the need for separate manual annotation campaigns or dedicated test vehicle runs.
Solution Approach 2:
The patent introduces probabilistic modeling and Bayesian inference as intermediary processes that bridge raw GPS trajectory data and accurate intersection maps. These computational methods automatically extract intersection structures from noisy vehicle track data, replacing manual annotation efforts while maintaining high map accuracy.
2Ease of manufacture
If simple road representations are used, then map data processing is easier, but lane-level and intersection information is insufficient for advanced driver safety systems
Solution Approach 1:
The patent segments the road network into discrete lane-level components and intersection structures. By representing roads as collections of lane segments connected at intersections with explicit turn lanes and valid path definitions, the system maintains detailed information while organizing data in a structured format suitable for processing by advanced driver safety systems.
Data Source
AI summary
A method of modeling an intersection structure of a roadway. The method includes receiving a first data set including road lane information, and receiving a second data set including vehicle trajectory information for an intersection structure of a roadway. The method includes determining lane node locations from at least one of the first and second data sets. A set of potential links between the lane node locations may be compiled. The method may further include assessing, for each link, a probability that the link is a valid link, and assigning each link with a probability value. The links may be filtered based on a predetermined threshold probability value and a set of valid links is generated. A model of the intersection structure is created based on the set of valid links.