In-Process Surface Normal Estimation for Raster Scanned Point Cloud Data

A method for in-process surface normal estimation from point cloud data is presented. The method enables surface normal estimation immediately after coordinates of points are measured. Such an approach allows in-process computational registration, used for collision and occlusion avoidance during dimensional inspection with high-precision point-based range sensors. The most commonly used sensor path for inspection with high-precision point-based range sensors is a raster scan path. A novel neighborhood identification approach for raster scanned point cloud data is presented. Quadratic polynomials are used to model the local geometry of the surface, from which the surface normal is estimated for the point. Implementation of the method through simulations and on a real part shows the normal estimation error to be within 0.1°.