Map Data Fusion Using Registration and Likelihood Alignment
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing map data for autonomous vehicles faces challenges such as calibration issues with on-board cameras, GPS bias and noise, lack of lane line attributes in telemetry-based data, precision limitations in aerial data, and time-consuming updates in high-definition data, necessitating an improved approach for procuring and fusing map data.
Innovation Solution
A system that uses central computers to align and fuse multiple versions of map data, including GPS, image, and telemetry data, through a map-matching registration algorithm and maximum likelihood estimation, creating fused map data with improved precision and accuracy by calculating probability distribution parameters and map fusion offsets.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If crowdsourced map data is used, then real-time information and lane line attributes are provided, but calibration issues with on-board camera and GPS bias introduce errors
Solution Approach 1:
The patent combines multiple map data sources (crowdsourced data, aerial data, and ground truth data) into a unified fused map. By merging these diverse sources with different characteristics, the system leverages the real-time updates from crowdsourced data while compensating for their accuracy limitations using the precision of ground truth data, thus resolving the contradiction between update speed and accuracy.
Solution Approach 2:
The patent introduces ground truth map data as an intermediary reference to correct and validate crowdsourced map data. This intermediary serves as a mediator that transfers accurate geometric information from survey-grade sources to the crowdsourced data, improving overall accuracy without sacrificing the real-time update capability of the crowdsourced source.
2Measurement precision
If aerial map data is used, then high precision is provided, but lane line attributes are inconsistent and creation is time-consuming
Solution Approach 1:
The patent segments the map data fusion process into distinct components: using aerial data primarily for high-precision geometric positioning while separately acquiring lane line attributes from crowdsourced data. This segmentation allows each data source to contribute its strengths without being constrained by the weaknesses of the other, achieving both precision and efficiency.
Solution Approach 2:
The fused map system achieves multi-functionality by integrating different data sources that serve different purposes: aerial data provides precise geospatial framework, while crowdsourced data supplies lane line attributes and real-time updates. This universal approach allows the system to simultaneously achieve high precision and fast creation/update rates.
3Measurement precision
If high-definition map data is used, then high precision and lane line attributes are provided, but updates are time-consuming
Solution Approach 1:
The patent implements a periodic update strategy where the complete high-definition map is created less frequently using ground truth data, while intermediate updates are performed more frequently using faster crowdsourced data sources. This periodic approach maintains high precision through periodic ground truth calibration while reducing average update time through interim crowdsourced updates.
Solution Approach 2:
The system applies partial action by using ground truth data only for critical geometric corrections rather than complete remapping, and using crowdsourced data for routine attribute updates. This selective application of different data sources based on update needs reduces overall processing time while maintaining necessary precision.
Data Source
AI summary
A system for fusing two or more versions of map data together includes one or more central computers that receive road network data representing a road network for a predefined geofenced area. The central computers compute a plurality of points that are each positioned at a predetermined distance from one another. The central computers create a plurality of bounding boxes for the road network based on the plurality of points and create a set of closest matched map data points for each bounding box that is part of the road network by executing a map-matching registration algorithm to align the two or more versions of map data with one another. The central computers execute a maximum likelihood estimation algorithm to determine probability distribution parameters of the set of closest matched map data points compared to the ground truth map data.