Loop-Partitioned Point Cloud Registration for GPS-Denied HD Mapping
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current point cloud registration algorithms for autonomous driving vehicles rely heavily on GPS signals, which can be unreliable in areas with tall buildings or dense forests, and are computationally complex, especially for large map areas, leading to inaccuracies and ghost image issues in high-definition 3D point cloud maps.
Innovation Solution
A system that partitions large map areas into smaller sections using topology information, allowing for parallel computation and improved registration of point clouds, even with weak or no GPS signals, by transforming point clouds from a relative to an absolute coordinate system and using a pose graph algorithm for merging, while also synchronizing RGB images with point clouds to enhance registration accuracy.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If GPS signal is used for point cloud registration, then localization is achieved, but accuracy deteriorates in areas with tall buildings or dense forests due to signal bouncing or failure
Solution Approach 1:
The patent introduces an intermediary system consisting of a portable computing device and a server that facilitates point cloud registration without relying on GPS signals. The portable device captures images and generates point clouds, which are then processed by the server to create location information through comparison with map data, serving as a mediator between the vehicle and the environment for accurate localization in GPS-denied areas
Solution Approach 2:
The patent replaces the GPS-based mechanical/signal system with an image-based visual odometry system. Instead of using radio frequency signals from GPS satellites, the system uses cameras to capture visual features and generate point clouds, substituting the signal-based localization mechanism with a vision-based approach that is not affected by tall buildings or dense forests
2Manufacturing precision
If conventional point cloud registration algorithms are used, then map construction is achieved, but computational complexity increases for large map areas
Solution Approach 1:
The patent segments the point cloud registration process into distinct components: image capture by the portable device, initial processing to generate point clouds, server-based processing to compare with map data and refine location information, and feedback to the vehicle. This segmentation allows each component to be optimized independently and processed in parallel, reducing overall computational complexity for large map areas
Solution Approach 2:
The system performs preliminary actions by pre-storing high-definition map data on the server and pre-processing images to generate point clouds before the actual registration is needed. This preliminary preparation reduces the computational burden during real-time operation, as the system only needs to perform comparisons and refinements rather than complete processing from scratch
3Area of stationary object
If point clouds are spliced to create HD 3D map, then map coverage is improved, but ghost image issues arise reducing map accuracy
Solution Approach 1:
The patent implements a feedback mechanism where the server compares the point cloud generated from captured images with the pre-stored high-definition map data. This comparison provides feedback on the accuracy of the location information, allowing the system to identify and correct ghost image issues by refining the registration parameters and adjusting the splicing process to eliminate artifacts
Solution Approach 2:
The system dynamically adjusts the point cloud splicing process based on the comparison results and feedback from the server. Rather than using a static splicing approach, the system modifies registration parameters, transformation matrices, and splicing strategies in real-time to minimize ghost images and maintain high accuracy across large map areas
Data Source
Figure 1
Figure 2
Figure 3A
AI summary
A system for registration of point clouds for autonomous driving vehicles is provided. The system receives a number of point clouds and corresponding poses from the autonomous driving vehicles equipped with LIDAR sensors capturing point clouds of a navigable area to be mapped, where the point clouds correspond to a first coordinate system. The system partitions the point clouds and the corresponding poses into one or more loop partitions based on navigable loop information captured by the point clouds. For each of the loop partitions, the system applies an optimization model to point clouds corresponding to the loop partition to register the point clouds. The system merges the one or more loop partitions together using a pose graph algorithm, where the merged partitions of point clouds are utilized to perceive a driving environment surrounding the autonomous driving vehicles.