Lane Elevation Estimation Using Headings for 3D Road Maps
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing systems face challenges in generating accurate three-dimensional (3D) maps with precise lane elevation information, leading to increased complexity and cost, particularly when dealing with varying road geometries and insufficient sensor data, which compromises vehicle safety and reliability, especially for automated driving systems.
Innovation Solution
An estimation system infers lane elevation using headings and a weighted average of surrounding lane boundaries from sensor data, without manual annotation, to generate 3D maps by augmenting detected lane boundaries with vehicle headings and direction vectors, and estimating vertex elevations using vehicle poses and directional differences.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Manufacturing precision
If manual inputs are used to produce accurate maps of lanes with elevation, then manufacturing precision is improved, but device complexity increases
Solution Approach 1:
The system uses automated algorithms to infer elevation information from sensor data and map data without requiring manual annotation. The elevation inference module automatically processes sensor data, determines vehicle poses, and calculates elevations using weighted averages, enabling the system to generate accurate 3D maps autonomously
Solution Approach 2:
The patent replaces manual annotation processes with automated computational methods. Instead of relying on human operators to manually input elevation data, the system uses algorithms that process sensor data, determine vehicle poses, and calculate elevations through weighted averages, substituting mechanical/manual operations with automated computational systems
2Manufacturing precision
If manual annotation is used to annotate map data with elevation information, then manufacturing precision is improved, but loss of time increases
Solution Approach 1:
The system performs automated elevation inference using sensor data and map data without requiring manual annotation. The elevation inference module continuously processes available data to determine elevations, enabling rapid generation of accurate 3D maps without the time-consuming manual input process
Solution Approach 2:
The system collects and processes sensor data in real-time as vehicles traverse the environment, accumulating the information needed for elevation inference before map generation is required. This preliminary data collection and processing enables rapid generation of accurate 3D maps without time-consuming manual annotation
3Device complexity
If sensor data is used to generate maps, then device complexity is reduced, but measurement precision deteriorates
Solution Approach 1:
The system uses a weighted average calculation that incorporates elevation information from multiple vehicles and adjusts the weighting based on directional differences and spatial relationships. This feedback mechanism allows the system to refine elevation estimates from sensor data, improving precision while maintaining automated processing
Solution Approach 2:
The system processes and integrates data from multiple sources including sensor data from various vehicles, map data, and elevation information from different locations. By universally processing diverse data types through a unified algorithmic framework, the system achieves high elevation precision while maintaining relatively simple system architecture
Data Source
AI summary
System, methods, and other embodiments described herein relate to estimating lane elevation using headings for generating three-dimensional (3D) maps. In one embodiment, a method includes augmenting positions of detected lane boundaries with headings of a vehicle from sensor data. The method also includes adding the headings to vertices of two-dimensional (2D) lines using direction vectors between the vertices, wherein the 2D lines are directed towards the detected lane boundaries. The method also includes estimating elevations of the vertices including surrounding lane boundaries using a weighted average and vehicle poses from the sensor data, the surrounding lane boundaries having locations different than the positions and the weighted average factors directional differences between the vertices. The method also includes generating a 3D map of driving lanes having stacked roads identified using the elevations.