A Maximum Entropy Based Outlier Removal for Airborne LiDAR Point Clouds
Ge Jiang,Derek D. Lichti,Tiangang Yin,Wai Yeung Yan
DOI: https://doi.org/10.1109/jstars.2024.3478069
IF: 4.715
2024-10-29
IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing
Abstract:Airborne light detection and ranging (LiDAR) data often suffer from noisy returns hovering in empty space within the collected 3-D point clouds. This can be attributed to system-induced factors, such as timing jitter and range walk error, or instantaneous air conditions, such as smoke, rain, clouds, etc. These floating points are indeed outliers, which significantly affect the subsequent analytical processes. Though various point cloud denoising methods are proposed based on sparsity assumption and elevation, they are highly unlikely to remove both clustered and scattered noisy points, especially those located close to the point clouds. Meanwhile, the performance of existing methods does not perform well when noisy points appear close to the ground or on rugged terrain. Accordingly, we propose a maximum entropy based outlier removal (MEOR) method for airborne LiDAR point clouds. More specifically, the proposed method includes two stages, i.e., one global coarse outlier removal stage (MEOR-G) and the subsequent local refined outlier removal stage (MEOR-L). In each stage, the MEOR algorithm is exploited to 1) produce an elevation histogram for the point clouds, 2) search for the elevation threshold to distinguish noisy points and valid points, and 3) remove noisy points and preserve valid data points. We conduct several comprehensive experiments to compare the performance of our proposed MEOR against four other existing noisy point removal methods on four LiDAR datasets. The experimental results demonstrate that MEOR significantly outperforms four other denoising methods by simultaneously removing clustered and scattered noisy points and achieves an improvement by 0.126–99.815%, 0–100%, 0.001–8.454%, and 0.053–99.691% in terms of recall, precision, overall accuracy, and F1 score, respectively.
imaging science & photographic technology,remote sensing,engineering, electrical & electronic,geography, physical