3D Submap Reconstruction 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 inaccuracies in simultaneous localization and mapping (SLAM) methods.
Innovation Solution
A system and method that utilize a camera-based 3D submap and LiDAR-based global map for localization, involving voxelization, feature extraction, and classification, with data alignment and inertial navigation to achieve centimeter precision by constructing and refining feature correspondences between submaps and global maps.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If GPS is used for localization in urban environments, then localization information can be obtained, but signal availability is poor and accuracy is compromised due to multi-path errors
Solution Approach 1:
The patent introduces an intermediary system consisting of a camera-based submap and LiDAR-based global map to bridge the gap between GPS unavailability and accurate localization. The visual/LiDAR SLAM system acts as a mediator that provides localization information when GPS signals are unavailable or inaccurate in urban environments with multi-path errors
Solution Approach 2:
The patent replaces the GPS satellite-based electromagnetic signal system with a ground-based visual/LiDAR sensing system. By substituting the mechanical/optical sensing approach (camera and LiDAR) for the satellite signal approach, the system achieves reliable localization in urban environments where GPS fails
2Reliability
If visual SLAM is used to construct 3D submap, then localization can be achieved in GPS-denied environments, but drift accumulates and accuracy degrades over time
Solution Approach 1:
The patent merges two separate mapping systems: a camera-based visual SLAM system that constructs a 3D submap and a LiDAR-based system that constructs a global map. By combining these two systems and performing registration between them, the patent achieves both continuous localization availability and improved accuracy by preventing drift accumulation through periodic global map alignment
Solution Approach 2:
The patent implements a feedback mechanism where the registered global map provides correction information to the visual SLAM system. The global map serves as a reference that feedbacks drift correction to the submap system, preventing error accumulation and maintaining long-term localization accuracy
3Measurement precision
If feature extraction and matching is performed between submap and global map, then localization precision is improved, but computational complexity increases
Solution Approach 1:
The patent segments the feature extraction and matching process into distinct stages: extracting features from the 3D submap, extracting features from the global map, and then performing matching between corresponding features. This segmentation allows for optimized processing at each stage and reduces overall computational complexity by breaking down the complex task into manageable components
Solution Approach 2:
The patent performs preliminary actions by pre-constructing both the 3D submap and global map before the actual localization task. By having these maps pre-prepared with extracted features available, the system reduces real-time computational complexity during localization operations, as the heavy feature extraction work has already been completed in advance
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 by one or more autonomous vehicle driving modules execution of processing of images from a camera and data from a LiDAR using the following steps comprising: voxelizing a 3D submap and a global map into voxels; estimating distribution of 3D points within the voxels, using a probabilistic model; extracting features from the 3D submap and the global map; and classifying the extracted features into classes.