Normal Estimation and Surface Reconstruction of Large Point Clouds

Estimating normals for 3D point clouds and reconstructing a surface interpolating the points are important problems in Computational Geometry and Computer Graphics. Massive point clouds and surfaces with sharp features are active areas of research for these problems. This thesis provides a fast and accurate algorithm for normal estimation and surface reconstruction which can handle large datasets as well as sharp edges and corners. We were successfully able to compute accurate normals for all the points on a cube including corners and edges and reconstruct the cube. We use several techniques to make the implementation fast and external-memory efficient. The implementation uses multiple threads operating in parallel and hence performs faster on a multiprocessor system. We use a technique called fast projective clustering for fitting multiple planes through the neighborhood of a point. We use a sliding window type streaming algorithm that uses a dynamic data structure for nearest neighbor search. We also develop a simplification algorithm that handles large datasets with sharp edges and corners and enables us to render these datasets using in-core rendering softwares.