Weighted Least Squares for Visualization of Scanned Point Clouds

This paper describes several improvements to implicit surface fitting over point clouds. Noisy stamped part point cloud data gathered by both a laser digitizer scanning system and a stereo vision sensor is processed using a method based on the weighted least squares algorithm. Two kernel functions are proposed: one using an enhanced sphere-of-influence graph (SIG) method to determine nearest neighbours; and a second using an improved Gaussian kernel. Automatic bandwidth adjustment is implemented for both methods. A method of incorporating a scalar attribute at each data point, which can represent any desired quantity and relate it to the surface at that point, is introduced. In this case surface strain data was gathered and processed by the scanning system, with a resulting thickness strain measurement calculated at each point. This strain data was used to introduce a temperature plot style colour map over the surface, allowing for fast and simple analysis of the strain of the formed part.