LiDAR Ground Filtering and Point Clustering with Reduced Processing Power
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing LiDAR systems face challenges in efficiently filtering ground and non-ground points and clustering non-ground points into objects due to high computational requirements and lack of reproducibility, particularly with methods like RANSAC and k-means, which are slow and prone to randomness.
Innovation Solution
A method involving a processor that maps LiDAR data to a 2D matrix, uses a Kalman filter to identify ground points, and performs a range check to cluster non-ground points, allowing for efficient and reproducible processing without the need for powerful processors.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If conventional methods like RANSAC or k-means are used for ground filtering and clustering, then filtering and clustering can be performed, but processing time increases and processing power requirements increase
Solution Approach 1:
The LiDAR data processing is segmented into distinct phases: ground filtering using RANSAC to identify ground planes, followed by separate clustering of non-ground points using k-means. This segmentation allows each algorithm to focus on specific tasks, improving overall efficiency while maintaining accuracy.
Solution Approach 2:
Ground filtering is performed as a preliminary action before clustering. By first identifying and removing ground points using RANSAC, the subsequent k-means clustering operates only on non-ground points, significantly reducing the data volume and processing time required for object detection.
2Reliability
If conventional methods like RANSAC or k-means are used for ground filtering and clustering, then filtering and clustering can be performed, but processing power requirements increase
Solution Approach 1:
The processing workload is segmented into two distinct stages with different computational requirements. The RANSAC ground filtering stage handles plane detection, while the k-means clustering stage handles object formation on reduced datasets, optimizing power usage for each task.
Solution Approach 2:
By performing ground filtering as a preliminary action that removes the majority of ground points before clustering, the system reduces the input data size for the computationally intensive k-means algorithm, thereby reducing overall processing power requirements while maintaining clustering accuracy.
3Reliability
If RANSAC or k-means methods are used, then ground filtering and clustering can be achieved, but the results are prone to randomness and lack reproducibility
Solution Approach 1:
The system implements feedback mechanisms where clustering results are evaluated against expected object characteristics, and ground filtering results are validated before proceeding to clustering. This feedback loop helps ensure consistent, reproducible results across different runs.
Solution Approach 2:
The system carefully controls and optimizes parameters for both RANSAC (such as number of iterations, inlier threshold) and k-means (such as number of clusters, initialization method) to reduce randomness. By stabilizing these parameters through systematic selection, the system improves reproducibility while maintaining the effectiveness of these probabilistic algorithms.
Data Source
Figure 1
Figure 2
Figure 3
AI summary
Systems and methods for a processor in connection with a light detection and ranging (LiDAR) sensor to cluster data points in a LiDAR data set using reduced processing power are described. The processor can receive a LiDAR dataset including a plurality of data points representing a plurality of points in a light detection and ranging (LiDAR) point cloud generated by a LiDAR device. For a group of data points having the same azimuth angle among the plurality of data points, the processor can, iteratively from a lowest elevation angle to a highest elevation angle, determining a search area of a specific data point, perform a range check on data points in the search area under a predefined sequence to identify neighbors of the specific data point, and cluster the specific data point with neighbors identified in the search area.