Point cloud denoising by reduction of ranging error in laser incidence direction

Laser scanning technology is gradually being used in tunnel 3d modeling and deformation monitoring to assist in tunnel construction and operation and maintenance management. After laser scanning tunnels, point cloud denoising is an important part of point cloud processing. Nevertheless, current statistical-based point cloud denoising algorithms can cause some point cloud information loss, resulting in voids or distortions in the tunnel monitoring area. In this paper, the characteristics of point cloud noise are revealed by analyzing the plane data set obtained from the experiment. And according to the point cloud noise characteristics, a new point cloud denoising algorithm is proposed from the perspective of point cloud noise generation rather than a statistical calculation method. It is found that point cloud noise is mainly caused by the ranging error along the laser incidence direction. The point cloud corresponds to a more consistent performance of the laser incidence angle, which has less influence on the point cloud noise. The point cloud denoising reduces the ranging error along the laser incidence direction. This denoising algorithm, derived from the principle of point cloud noise generation instead of the conventional statistical noise reduction method, provides a noise reduction perspective for point cloud noise reduction. Furthermore, this denoising algorithm substitutes the original sampling point with a point located on its laser ray. This point is specifically identified as the intersection of the laser ray of the sampling point and the plane fitted with its neighboring points. This algorithm employs the newly sampled point in place of the original point cloud, thereby preserving the information within the point cloud. It is also capable of executing multiple noise reduction procedures on the point cloud, with the most significant effects observed in the first three iterations. This method is particularly appropriate for measurements that demand high sensitivity to original data, such as tunnel deformation monitoring and heritage detection.