Point cloud denoising by robust PCA dimension reduction

• Main Author: Guo, Yu
• Other Authors: Wen Bihan
• Format: Thesis-Master by Coursework
• Language: English
• Published: Nanyang Technological University 2024
• Subjects: Engineering
• Online Access: https://hdl.handle.net/10356/178730
• Institution: Nanyang Technological University

Lidar sensors are often used to scan point cloud data, but the signals it sends are often negatively affected by atmospheric particles, light scattering and other influencing factors during transmission, resulting in noise in point cloud images. In this dissertation, advanced point cloud data denoising and dimensionality reduction technology is studied, which is a density-based spatial clustering of application with noise (DBSCAN) algorithm based on robust principal component analysis (RPCA), aiming to improve the accuracy and efficiency of high point cloud data processing. Different from directly processing 3D point clouds, we reduce the dimension of data to 2D point clouds through RPCA technology, which effectively reduces the impact of noise and outliers, and also increases robustness compared with traditional PCA, while retaining the characteristics of less information consumption. This is combined with an adaptive clustering mechanism that dynamically adjusts parameters based on local data features, which has been shown to significantly improve the preservation of basic data features while removing redundant and noisy data. Point cloud data will be restored to 3D space after clustering and noise reduction. Experimental results of simulated and real data sets show that the proposed method not only reduces noise levels, but also preserves key structural details more efficiently than traditional methods. The combination of RPCA and DBSCAN algorithms called robust PCA adaptive clustering (RPCAAC) was validated through extensive computational performance analysis, showing substantial improvements in processing efficiency and data fidelity.