Filtering of remote sensing point clouds using fuzzy C-means clustering

Abstract

Standard filtering techniques perform well on digital surface models (DSMs) generated from light detection and ranging (LiDAR) data. However, these techniques have to be tested and evaluated using point clouds obtained by matching of stereo satellite imagery. This paper proposes a new iterative filtering technique based on fuzzy c-means (FCM) clustering. This method is composed of three main steps: (1) a DSM is generated from GeoEye-1 stereo pair imagery, (2) the generated DSM is then reshaped and applied as input data for a FCM clustering process to separate terrain and non-terrain points, and (3) terrain points are then interpolated into a grid digital terrain model (DTM). An urban test area with distinct land use/cover classes covering the north-east part of Cairo city in Egypt has been selected. To evaluate the performance of the proposed method, the filtered DTM was compared against reference data that was generated manually, and type I, type II, and total errors were estimated. Compared with the most commonly used filtering method, progressive triangular irregular network densification (PTD), the proposed approach has identified terrain points much closer to the reference DTM with less variable accuracies. The FCM filter has reduced the mean errors by 2.06, 0.74, and 0.34% for type I, type II, and total errors, respectively. On the other hand, the mean standard deviation (SD) of the differences between the obtained and reference DTMs has been reduced by 0.38 m.