基于法向偏差的隧道点云去噪算法

Tunnel point clouds denoising algorithm based on normal deviation

针对隧道类似圆柱的形态特征以及三维点云法向对噪声的敏感性,设计一种剔除由通行车辆和隧道内壁悬挂物所造成噪声点的方法。该方法利用点云的大量法向量鲁棒地估计出精确的隧道轴向,并根据点云法向与隧道轴向的偏差识别出可靠的隧道表面点,然后参照可靠点完成噪声点的进一步确认。利用仿真数据和真实的高速公路隧道扫描点云实验结果证明该方法的有效性和精确性。

Considering the shape characteristics of tunnel similar to cylinder, this paper presents a method of eliminating the noises from tunnel laser scanned points caused by passing vehicles and tunnel attached facilities, which utilizes the fact that normal estimation of three-dimensional point clouds is sensitive to noise. This method accurately predicts the tunnel axis direction from the large volume of normal vectors so that reliable tunnel surface points can be identified according to their normal angles with the tunnel axis direction, and then based on the distances to the reliable data, noise points are progressively segmented out from the laser scanned raw data. The proposed method has been validated on both synthetic data and real point clouds scanning from a highway tunnel, and the experimental result demonstrates that it can effectively preserve the geometric details of tunnel surfaces while removing the two types of noises.