Segmented Point Cloud Registration for Lane-Level HD Mapping
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional maps have low accuracy and cannot provide lane-level route planning, limiting the effectiveness of autonomous driving systems, while accurately registering point cloud data is crucial for constructing high-definition maps.
Innovation Solution
A method for point cloud registration involving dividing target point cloud data into sets, determining coincidence degrees, and using a registration matrix for precise alignment of point cloud sets, including filtering and smoothing processes to enhance accuracy.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If conventional map construction methods are used, then the mapping process is simple, but the map accuracy is low and cannot provide lane-level route planning
Solution Approach 1:
The patent divides the target point cloud data into multiple point cloud sets based on time intervals, and further segments each set into multiple blocks. This segmentation enables efficient processing and registration of large-scale point cloud data, resolving the contradiction between achieving high map accuracy and managing registration process complexity.
Solution Approach 2:
The patent performs preliminary actions by pre-dividing point cloud data into structured sets and blocks, and pre-calculating coincidence degrees between blocks before actual registration. This preliminary organization and preprocessing reduce the complexity of the main registration process while ensuring high accuracy in map construction.
2Measurement precision
If point cloud data is processed without division, then the processing is simpler, but the registration accuracy decreases
Solution Approach 1:
The patent divides target point cloud data into multiple point cloud sets according to time intervals, and each set is further divided into multiple blocks. This multi-level segmentation structure enables precise registration by comparing corresponding blocks between different time points, significantly improving registration accuracy while managing data processing complexity through systematic organization.
Solution Approach 2:
The patent introduces a temporal dimension by dividing point cloud data according to time intervals, creating point cloud sets at different time points. This temporal segmentation allows for more accurate registration by capturing changes over time, while the block-level division within each time set manages the computational complexity of processing the expanded data structure.
3Reliability
If all point cloud sets are registered pairwise, then the registration completeness is high, but the computational time increases
Solution Approach 1:
The patent divides point cloud data into multiple sets and further segments each set into blocks, then performs registration by comparing corresponding blocks between sets. This block-level comparison approach ensures registration completeness by systematically covering all relevant point cloud regions, while significantly reducing computational time compared to registering entire point cloud sets pairwise.
Solution Approach 2:
The patent performs registration on representative blocks within point cloud sets rather than exhaustively registering all possible pairs of point cloud sets. This partial action approach maintains registration completeness by ensuring all critical regions are covered through the block-level division, while dramatically reducing the computational burden of exhaustive pairwise registration.
Data Source
AI summary
A point cloud registration method, apparatus, device, and storage medium are provided. The method includes: acquiring target point cloud data; dividing the target point cloud data into a plurality of point cloud sets; determining a coincidence degree between every two point cloud sets and determining a fixed point cloud set and a registration point cloud set from two point cloud sets with a coincidence degree between the two point cloud sets being greater than a preset threshold; determining a target registration matrix between the fixed point cloud set and the registration point cloud set; and performing registration of the fixed point cloud set with the registration point cloud set according to the target registration matrix.