Point Cloud Registration via Clustered ICP for HD Map Convergence
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current point cloud registration algorithms for autonomous driving vehicles are hindered by dependencies on GPS signals, which can be error-prone in urban environments and tunnels, and are computationally complex, especially when dealing with large map areas, limiting the accuracy and efficiency of high-definition map construction.
Innovation Solution
A method is introduced to partition the map region for regional iterative closest point (ICP) algorithms, determining the frame pair size and dividing it into clusters if it exceeds a memory-based threshold, allowing for parallel processing and reducing computational complexity by solving bundle adjustment equations separately within each cluster.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Device complexity
If GPS signals are used for vehicle localization in point cloud registration, then the system can operate with simple infrastructure, but the localization accuracy deteriorates in urban environments and tunnels due to signal errors and multi-path effects
Solution Approach 1:
The patent segments the point cloud data into multiple sub-maps, each processed independently by different computation nodes. This segmentation allows the system to achieve high localization accuracy for each sub-map without requiring complex GPS infrastructure, as each node processes a manageable portion of the data with sufficient computational resources.
Solution Approach 2:
The patent introduces an intermediary coordinate system transformation mechanism that bridges the gap between GPS-based localization and precise point cloud alignment. By using intermediate coordinate frames and transformation matrices, the system achieves accurate localization even when GPS signals are degraded, without requiring complete replacement of the GPS infrastructure.
2Productivity
If the map area is divided into sub-maps for parallel processing, then computational complexity is reduced, but the sub-map size is limited by computation node memory capacity
Solution Approach 1:
The patent implements dynamic sub-map partitioning where the size and boundaries of sub-maps are adjusted based on the computational capacity and memory availability of each computation node. This dynamic adaptation allows the system to maximize processing parallelism while ensuring that each node receives a sub-map size appropriate for its resources, thereby maintaining both high productivity and flexible adaptability.
3Productivity
If regional ICP algorithm is used on computation nodes, then point cloud registration can be performed in parallel, but the registration accuracy is influenced by the memory size of computation nodes
Solution Approach 1:
The patent merges the results from multiple regional ICP registrations performed in parallel on different computation nodes. By combining the localized transformations from each sub-map processing node into a global coordinate system, the system achieves both the processing efficiency of parallel computation and the accuracy of comprehensive point cloud alignment, overcoming the memory limitations of individual nodes.
Data Source
AI summary
In one embodiment, a system and method for partitioning a region for point cloud registration of LIDAR poses of an autonomous driving vehicle (ADV) using a regional iterative closest point (ICP) algorithm is disclosed. The method determines the frame pair size of one or more pairs of related LIDAR poses of a region of an HD map to be constructed. If the frame pair size is greater than a threshold, the region is further divided into multiple clusters. The method may perform the ICP algorithm for each cluster. Inside a cluster, the ICP algorithm focuses on a partial subset of the decision variables and assumes the rest of the decision variables are fixed. To construct the HD map, the method may determine if the results of the ICP algorithms from the clusters converge. If the solutions converge, a solution to the point cloud registration for the region is found.