3D Submap and Global Map Alignment for Centimeter Precision Localization
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Autonomous vehicles face challenges in robust and accurate localization in urban environments due to GPS signal unavailability and multi-path errors, leading to significant drift in visual/LiDAR SLAM methods and limited decimeter-level accuracy.
Innovation Solution
A method and system that aligns a 3D submap with a LiDAR-based global map by extracting and classifying features, establishing correspondence, and iteratively refining the location using camera and LiDAR data, achieving centimeter-level precision through voxelization and probabilistic modeling.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If visual/LiDAR SLAM methods are used for localization, then GPS signal unavailability is addressed, but drift occurs and accuracy is limited to decimeter-level
Solution Approach 1:
The patent divides the localization problem into two segments: visual SLAM for short-term tracking and map alignment for long-term accuracy. The submap is constructed from camera images using visual SLAM, then aligned with the pre-built LiDAR global map to correct drift and achieve centimeter-level precision.
Solution Approach 2:
The patent merges two different sensing modalities (visual camera data and LiDAR range data) into a unified localization system. The visual submap and LiDAR global map are registered through feature matching and coordinate transformation, combining the advantages of both sensing approaches.
2Measurement precision
If feature extraction and matching is performed to improve localization accuracy, then computational complexity increases
Solution Approach 1:
The patent extracts features from specific local regions of interest in the submap and global map rather than processing the entire map uniformly. This localized feature extraction reduces computational load while maintaining accuracy in critical localization areas.
Solution Approach 2:
The patent performs preliminary feature extraction and classification before the actual alignment process. Features are pre-processed and organized into different categories, which simplifies the subsequent matching and correspondence establishment steps.
Data Source
AI summary
A method of localization for a non-transitory computer readable storage medium storing one or more programs is disclosed. The one or more programs comprise instructions, which when executed by a computing device, cause the computing device to perform utilizing one or more autonomous vehicle driving modules that execute processing of images from a camera and data from a LiDAR the following steps comprising: aligning a 3D submap with a global map; extracting features from the 3D submap and the global map; classifying the extracted features in classes; and establishing correspondence of features in a same class between the 3D submap and the global map.