Lane Marking Localization Without Longitudinal Jitter
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing localization methods for autonomous vehicles, such as point-to-point iterative closest point (ICP) and 3D generalized iterative closest point (GICP), struggle with accurate matching of LIDAR point cloud data to lane markings due to ambiguity in the longitudinal direction, leading to jitter and inconsistent results in feature-poor environments like those with only lane-line features.
Innovation Solution
An improved point-to-line ICP method that minimizes the influence of the longitudinal variable by matching query points to a perpendicular point on the line, rather than the closest sampled point, providing more reliable lateral localization in line-based environments.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If point-to-point ICP or 3D GICP methods are used for localization, then the system can process LIDAR point cloud data, but the localization accuracy deteriorates due to ambiguity in the longitudinal direction causing jitter
Solution Approach 1:
The patent extracts and removes the longitudinal variable from the point-to-line fitting process. By minimizing only the lateral distance between query points and the fitted line while excluding longitudinal displacement, the method eliminates the source of ambiguity and jitter, thereby improving both localization accuracy and consistency in lane-marking environments
Solution Approach 2:
The patent changes the optimization parameters of the fitting algorithm by modifying the objective function to minimize lateral distance only. This parameter change transforms the fitting criterion from a conventional point-to-point or point-to-line method that includes longitudinal components to a specialized method that exclusively optimizes lateral positioning, resolving the jitter issue
2Ease of manufacture
If conventional point-to-point ICP is used, then the algorithm is simple to implement, but the matching accuracy to lane markings deteriorates in feature-poor environments
Solution Approach 1:
The patent transitions from point-to-point matching to point-to-line fitting, changing the dimensional relationship between the query points and reference features. By fitting points to entire lane marking lines rather than individual points, the method leverages the linear structure of lane markings to improve matching accuracy while maintaining algorithmic simplicity through the use of standard least-squares fitting techniques
Data Source
AI summary
System, methods, and computer-readable media for localizing an autonomous vehicle to a location in a stored LIDAR map relative to lane markings labeled in the stored LIDAR map. The labeled lane marking may be associated with a dashed line, a solid line, a double line, or an unmarked line. The autonomous vehicle may fit the set of lane marking points with the labeled lane marking via a point-to-line solver that locates the set of lane marking points to the line, excluding a longitudinal variable. The lane matching localizing system utilities a point-to-line iterative closet point, wherein the matched point pairs are calculated by perpendicular point of line from the query point instead of finding the closest sampled point.