Denoising Method Based on Improved DBSCAN for Lidar Point Cloud

Lidar 3D point cloud data often contains significant noise, which affects the accuracy and reliability of subsequent data analysis. This study aims to propose an improved adaptive DBSCAN (Density-Based Spatial Clustering of Applications with Noise) algorithm to more effectively remove noise from LiDAR 3D point cloud data, ensuring high noise reduction accuracy and retention of original points, whereas maintaining good adaptability and robustness across different point cloud distributions and noise levels. The method determines the Eps range by fitting the inflection point of the distance matrix curve and identifies suitable parameters for point cloud data processing by finding the optimal Calinski-Harabasz index value. Experimental results show that this method achieves noise reduction accuracy and original point retention rates of over 90% when processing different types of point cloud data, whereas effectively preserving environmental and target information. This method addresses the limitations of adaptive parameter determination in existing DBSCAN-based denoising research by dynamically adjusting parameters according to the density characteristics of LiDAR point cloud data, meeting the needs of different point cloud distributions and noise levels.