A Singular Workflow for 3d Surface Reconstruction of Heavily Noisy Point Clouds

Surface reconstruction from low quality point clouds represents a common problem in most standard algorithms created for this purpose. Point clouds acquired using specialized devices, such as 3D scanners, or as outputs from structure from motion algorithms are usually flawed in that they contain a significant amount of noise and outliers, making the surface reconstruction process difficult, resulting in low quality surface estimation. The quality of the reconstructed mesh is directly proportional to the quality of the point cloud itself. This paper proposes a workflow for creating 3D surfaces from unstructured point clouds. The workflow takes an unstructured point cloud as input and, through four phases, automatically cleans up the point cloud data and creates a watertight surface reconstruction of the point cloud, all in a single, end-to-end workflow.