LiDAR Point Cloud Registration via Track Segmentation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing techniques for registering a LiDAR point cloud with respect to a known three-dimensional point cloud in SLAM applications are prone to poor convergence and long operation durations when the initial position accuracy is low, often resulting in incorrect local optimal solutions.
Innovation Solution
A point cloud processing system that includes a sensing information acquisition means for acquiring sensing information, including a sensing point cloud generated by a LiDAR apparatus mounted on a moving body, and a first registration means for registering the sensing point cloud with respect to a travel track point cloud associated with the moving body's travel track, within a travel environment point cloud.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If traditional SLAM registration techniques (ICP, NDT) are used to register LiDAR point cloud with known three-dimensional point cloud, then registration accuracy can be achieved, but operation duration becomes long and convergence becomes poor when initial position accuracy is low
Solution Approach 1:
The patent segments the large-scale travel environment point cloud into multiple travel track point clouds, each corresponding to a specific travel track. This segmentation reduces the search space for registration, enabling faster convergence while maintaining accuracy. The sensing point cloud is registered against the specific travel track point cloud rather than the entire environment, significantly reducing operation duration.
Solution Approach 2:
The patent performs preliminary classification of the travel environment point cloud into multiple travel track point clouds before registration. This preliminary organization creates ready-to-use reference data structures that enable rapid registration when the moving body needs to register its sensing point cloud, avoiding the need to process the entire point cloud during the registration operation.
2Reliability
If traditional SLAM registration techniques are used with poor initial position accuracy, then registration may converge to wrong local optimal solutions, but avoiding this requires more complex preprocessing
Solution Approach 1:
The patent applies local quality by creating specific travel track point clouds that are tailored to each individual travel track. Each travel track point cloud contains geometric features specific to its corresponding track, enabling the registration algorithm to converge to the correct global optimum even when initial position accuracy is poor. This localized approach ensures reliability without requiring complex global preprocessing.
3Area of stationary object
If the entire travel environment point cloud is used for registration, then comprehensive coverage is achieved, but registration time increases significantly
Solution Approach 1:
The patent segments the large-scale travel environment point cloud into multiple smaller travel track point clouds, each covering a specific travel track. This segmentation maintains comprehensive coverage of the entire environment while enabling rapid registration by limiting the search space to only the relevant travel track. The system achieves both wide coverage and high registration speed through this segmented approach.
Data Source
AI summary
Provided is a point cloud processing system including: a sensing information acquisition means for acquiring sensing information including a sensing point cloud generated by a sensing means, which is mounted on a moving body traveling on any one of a plurality of travel tracks extending in parallel to one another, for sensing an area ahead in a heading direction of the moving body, and track identification information identifying a travel track on which the moving body travels while performing sensing; and a first registration means for registering the sensing point cloud with respect to a travel track point cloud associated with the travel track on which the moving body travels while performing sensing, among a travel environment point cloud being a known point cloud of a travel environment including the plurality of travel tracks.