A Radiometric Correction based Optical Modeling Approach to Removing Reflection Noise in TLS Point Clouds of Urban Scenes

TL;DR

This work tackles reflection noise in terrestrial LiDAR (TLS) urban point clouds by introducing a radiometric-correction–driven approach that directly detects reflective regions in 3D space and estimates reflective planes. It then removes virtual points using a reflection-invariant RE-LSFH descriptor paired with a Hausdorff-distance based similarity measure, enabling robust discrimination even under ghosting and deformation. The authors validate on the 3DRN dataset, showing substantial gains in precision/recall for reflective regions, improved outlier detection, and higher overall accuracy compared to state-of-the-art baselines. The approach improves TLS-based urban scene modeling by effectively mitigating reflective noise while preserving real geometry, and it introduces a publicly available 3DRN dataset to benchmark virtual-point removal techniques.

Abstract

Point clouds are vital in computer vision tasks such as 3D reconstruction, autonomous driving, and robotics. However, TLS-acquired point clouds often contain virtual points from reflective surfaces, causing disruptions. This study presents a reflection noise elimination algorithm for TLS point clouds. Our innovative reflection plane detection algorithm, based on geometry-optical models and physical properties, identifies and categorizes reflection points per optical reflection theory. We've adapted the LSFH feature descriptor to retain reflection features, mitigating interference from symmetrical architectural structures. By incorporating the Hausdorff feature distance, the algorithm enhances resilience to ghosting and deformation, improving virtual point detection accuracy. Extensive experiments on the 3DRN benchmark dataset, featuring diverse urban environments with virtual TLS reflection noise, show our algorithm improves precision and recall rates for 3D points in reflective regions by 57.03\% and 31.80\%, respectively. Our method achieves a 9.17\% better outlier detection rate and 5.65\% higher accuracy than leading methods. Access the 3DRN dataset at (https://github.com/Tsuiky/3DRN).

Figures (11)

• Figure 1: The principle of reflection in 3D point clouds captured by TLS. Laser beams emitted by TLS bounce off glass surfaces, producing 3D virtual points in the captured point cloud that do not exist in the real world.
• Figure 2: Illustration of reflection noise in RGB panorama image and TLS point cloud. (a) Reflective noise in an RGB panorama image. (b) Reflective noise in a large-scale TLS point cloud is structured to exhibit geometric shapes and semantic information similar to those of real points.
• Figure 3: The workflow of the proposed method features a reflective region detection module for precise estimation of reflective surfaces and a virtual point detection and removal module aimed at maximizing the removal of virtual points while maintaining real ones.
• Figure 4: Comparison of the proposed reflective area detection module (top) with the 2D projection-based approach (bottom). By taking full advantage of optical properties, the proposed module can directly extract dense points of possible reflective areas from 3D space, enabling accurate reflective plane estimation.
• Figure 5: Construction of the RE-LSFH feature descriptor. (a) Angle of deviation between the laser path and the normals. (b) Projection distance from the query point to the neighboring radii along the laser path.