The International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences
Download
Publications Copernicus
Download
Citation
Articles | Volume XLII-1
https://doi.org/10.5194/isprs-archives-XLII-1-87-2018
https://doi.org/10.5194/isprs-archives-XLII-1-87-2018
26 Sep 2018
 | 26 Sep 2018

STATISTICAL OUTLIER DETECTION METHOD FOR AIRBORNE LIDAR DATA

A. C. Carrilho, M. Galo, and R. C. Santos

Keywords: LiDAR data, Outlier detection, Point cloud, Frequency filter, Histogram analysis

Abstract. Sampling the Earth’s surface using airborne LASER scanning (ALS) systems suffers from several factors inherent to the LASER system itself as well as external factors, such as the presence of particles in the atmosphere, and/or multi-path returns due to reflections. The resulting point cloud may therefore contain some outliers and removing them is an important (and difficult) step for all subsequent processes that use this kind of data as input. In the literature, there are several approaches for outlier removal, some of which require external information, such as spatial frequency characteristics or presume parametric mathematical models for surface fitting. A limitation on the height histogram filtering approach was identified from the literature review: outliers that lie within the ground elevation difference might not be detected. To overcome such a limitation, this paper proposes an adaptive alternative based on point cloud cell subdivision. Instead of computing a single histogram for the whole dataset, the method applies the filtering to smaller patches, in which the ground elevation difference can be ignored. A study area was filtered, and the results were compared quantitatively with two other methods implemented in both free and commercial software packages. The reference data was generated manually in order to provide useful quality measures. The experiment shows that none of the tested filters was able to reach a level of complete automation, therefore manual corrections by the operator are still necessary.