Accurate Normal Calculating and Surface Smoothing of Laser-Scanned Point Clouds

In order to preserve feature while smoothing, a new bilateral filter with constraints is proposed on the basis of precise calculation of surface normal vectors. To get accurate normal vectors calculating results, a pre-sampling method is organized firstly to increase normal pre-estimating accuracy considering the characteristics of laser-scanned point clouds data. Robust least-squares estimation method is applied to increase normal estimation accuracy. And a new bilateral filter based on mean square error and distance of neighbor points is constructed. To deal with the problems of traditional bilateral filter used in smoothing, a new concept of "noise degree" is introduced. Based on noise degree and distance of neighbor points, a bilateral filter with constraints of normal vectors intersection angles and prediction distance of neighbor points is constructed. Experimental results demonstrate the feature-preserving ability and high efficiency of the algorithm.