Date of Award
Doctor of Philosophy (PhD)
Modeling Simul & Visual Engineering
Ahmed K. Noor
Point clouds and 3D meshes are widely used in numerous applications ranging from games to virtual reality to autonomous vehicles. This dissertation proposes several approaches for noise removal and calibration of noisy point cloud data and 3D mesh sharpening methods. Order statistic filters have been proven to be very successful in image processing and other domains as well. Different variations of order statistics filters originally proposed for image processing are extended to point cloud filtering in this dissertation. A brand-new adaptive vector median is proposed in this dissertation for removing noise and outliers from noisy point cloud data.
The major contributions of this research lie in four aspects: 1) Four order statistic algorithms are extended, and one adaptive filtering method is proposed for the noisy point cloud with improved results such as preserving significant features. These methods are applied to standard models as well as synthetic models, and real scenes, 2) A hardware acceleration of the proposed method using Microsoft parallel pattern library for filtering point clouds is implemented using multicore processors, 3) A new method for aerial LIDAR data filtering is proposed. The objective is to develop a method to enable automatic extraction of ground points from aerial LIDAR data with minimal human intervention, and 4) A novel method for mesh color sharpening using the discrete Laplace-Beltrami operator is proposed.
Median and order statistics-based filters are widely used in signal processing and image processing because they can easily remove outlier noise and preserve important features. This dissertation demonstrates a wide range of results with median filter, vector median filter, fuzzy vector median filter, adaptive mean, adaptive median, and adaptive vector median filter on point cloud data. The experiments show that large-scale noise is removed while preserving important features of the point cloud with reasonable computation time. Quantitative criteria (e.g., complexity, Hausdorff distance, and the root mean squared error (RMSE)), as well as qualitative criteria (e.g., the perceived visual quality of the processed point cloud), are employed to assess the performance of the filters in various cases corrupted by different noisy models. The adaptive vector median is further optimized for denoising or ground filtering aerial LIDAR data point cloud. The adaptive vector median is also accelerated on multi-core CPUs using Microsoft Parallel Patterns Library. In addition, this dissertation presents a new method for mesh color sharpening using the discrete Laplace-Beltrami operator, which is an approximation of second order derivatives on irregular 3D meshes. The one-ring neighborhood is utilized to compute the Laplace-Beltrami operator. The color for each vertex is updated by adding the Laplace-Beltrami operator of the vertex color weighted by a factor to its original value. Different discretizations of the Laplace-Beltrami operator have been proposed for geometrical processing of 3D meshes. This work utilizes several discretizations of the Laplace-Beltrami operator for sharpening 3D mesh colors and compares their performance. Experimental results demonstrated the effectiveness of the proposed algorithms.
"Adaptive Methods for Point Cloud and Mesh Processing"
(2018). Doctor of Philosophy (PhD), Dissertation, Modeling Simul & Visual Engineering, Old Dominion University, DOI: 10.25777/ttaf-b623