Improving data reduction for 3D shape preserving

Abstract

A novel 3D shape preserving data reduction technique for managing the amount of data acquired by laser scanning is presented that overcomes the shortcomings of existing filter-based methods. The technique is based on a discrete Gaussian image of the scanned points which is obtained by estimating surface normals and projecting them into a Gaussian sphere. The discrete Gaussian image is then used to partition the points into cells. In each cell, a reference point and its neighbours are used to determine the cell representative point and all other points are removed. The performance of the proposed method is illustrated using a range of point clouds scanned from typical engineering surfaces.