LiDAR Point Cloud Processing With Adaptive Kernel Pose Adjustment
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing LiDAR data processing methods struggle to achieve both local consistency and global accuracy in constructing point clouds due to noise, measurement errors, and varying sensor poses, particularly in complex environments.
Innovation Solution
A system and method combining spatial smoothing and pose adjustment modules, utilizing a polynomial smoothing kernel and iterative kernel size adjustment to refine LiDAR data, incorporating Principal Component Analysis (PCA) for sampling and Gaussian radial weight functions to enhance local consistency and global accuracy.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Stability of the object's composition
If spatial smoothing with a fixed kernel size is applied to LiDAR data, then local consistency is improved, but global accuracy deteriorates due to inability to adapt to varying scales in different regions
Solution Approach 1:
The patent applies dynamic kernel size adjustment where the kernel size is no longer fixed but adapts iteratively based on the data being processed. The kernel size is updated in each iteration based on the convergence of pose adjustment, allowing the smoothing operation to adapt to different spatial scales in different regions of the point cloud, thereby resolving the contradiction between local consistency and global accuracy.
Solution Approach 2:
The patent changes the parameter of kernel size from a static value to a dynamic parameter that evolves during processing. By iteratively updating the kernel size based on pose adjustment convergence, the system can use larger kernels for regions requiring more smoothing and smaller kernels for regions requiring finer detail preservation, thus achieving both local consistency and global accuracy.
2Measurement precision
If iterative pose adjustment is performed to improve global accuracy, then computational time increases due to repeated kernel size updates and convergence checks
Solution Approach 1:
The patent implements a feedback mechanism where the convergence of pose adjustment is continuously monitored and used to control the iterative process. The kernel size is updated only when convergence criteria are met, and the iteration stops when further adjustments no longer improve accuracy beyond the threshold. This feedback-based approach minimizes unnecessary computations while ensuring global accuracy is achieved.
Solution Approach 2:
The patent applies partial iteration by performing pose adjustment and kernel size updates only as many times as necessary to achieve convergence. Rather than performing a fixed large number of iterations, the system performs the minimum number of iterations required to meet the convergence criterion, thus reducing computational time while maintaining global accuracy.
3Productivity
If PCA sampling is applied to reduce point cloud density, then processing speed is improved, but loss of fine geometric details occurs
Solution Approach 1:
The patent applies segmentation by dividing the point cloud processing into different stages: initial PCA sampling for rapid processing and coarse feature extraction, followed by iterative spatial smoothing with adaptive kernel sizes for refined detail preservation. This segmented approach allows different processing strategies to be applied to different aspects of the data, achieving both speed and precision.
Solution Approach 2:
The patent applies local quality by using adaptive kernel sizes that vary across different regions of the point cloud. Regions with fine geometric details receive smaller kernel sizes to preserve detail, while regions with smoother surfaces receive larger kernel sizes for efficient processing. This local differentiation allows the system to maintain geometric precision where needed while achieving high processing speed overall.
Data Source
AI summary
Systems and methods for LiDAR point cloud processing are disclosed that can include a point cloud data acquisition module configured to receive point cloud data, a spatial smoothing module configured to generate a fitted surface from the point cloud data using a predetermined kernel size, where the fitted surface provides an estimation of locations of points in the point cloud data, a pose adjustment module configured to perform pose adjustment on a LIDAR frame of the point cloud data by calculating an error between points on the fitted surface and corresponding data points in the point cloud data, and adjusting a pose of the LiDAR frame to minimize the error, and a convergence determination module configured to assess a convergence of the pose adjustment by comparing a change in pose adjustment to a threshold value, where if the change exceeds the threshold value, the kernel size is updated.