Filtering Airborne Laser Scanning Data with Morphological Methods

Filtering methods based on morphological operations have been developed in some previous studies. The biggest challenge for these methods is how to keep the terrain features unchanged while using large window sizes for the morphological opening. Zhang et al. (2003) tried to achieve this goal, but their method required the assumption that the slope is constant. This paper presents a new method to achieve this goal without such restrictions, and methods for filling missing data and removing outliers are proposed. The experimental test results using the ISPRS Commission III/WG3 dataset show that this method performs well for most sites, except those with missing data due to the lack of overlap between swaths. This method also shows encouraging results for laser data with low pulse density. Introduction Airborne laser scanning (ALS) is gaining popularity in various environmental applications, ranging from DEM mapping, transportation, and urban studies to forest management, hydrology, and ecology (Flood and Gutelius, 1997; Chen et al., 2006). Compared with digital photogrammetry (Gong et al., 2000; Sheng et al., 2001; Gong et al., 2002) and radar interferometry (Hoekman and Varekamp, 1998), laser altimetry has the advantage of recording the elevation of earth surface directly. Nevertheless, there is a great need for efficient data processing methods (Axelsson, 1999). In particular, filtering, the abstraction of bare earth from ALS points, is a crucial procedure for ALS data processing (Sithole and Vosselman, 2004). It, with quality control, generally consumes an estimated 60 to 80 percent of processing time (Flood, 2001). However, the details of filtering Filtering Airborne Laser Scanning Data with Morphological Methods Qi Chen, Peng Gong, Dennis Baldocchi, and Gengxin Xie algorithms were seldom reported due to the tendency of some commercial and academic practitioners to keep their work proprietary (Haugerud and Harding, 2001; Huising and Gomes-Pereira, 1998; Sithole and Vosselman, 2004). Sithole and Vosselman (2004) divided current filtering algorithms into four categories including slope-based, blockminimum, surface-based, and clustering/segmentation methods, among which the surface-based method is widely used. The idea of surface-based methods is to create a surface with a corresponding buffer zone above it, and the buffer zone defines the region in 3D space where terrain points are expected to reside (Sithole and Vosselman, 2004). The key of this method is to create a surface approximating the bare earth. Depending on the means of creating the surface, surface-based filtering methods can be further divided into the following two subcategories: 1. Interpolation-based Methods: Kraus and Pfeifer (1998) proposed an algorithm to iteratively approximate the ground using weighted linear least squares interpolation. Since terrain points usually have negative residuals and nonterrain points have positive ones, a weight function was designed to assign high weight to the points with negative residuals. This algorithm was extended by incorporating the hierarchical approach (Pfeifer et al., 2001), and it was found that the hierarchical interpolation can improve the filter result and speed up the computation. Lee and Younan (2003) improved Kraus and Pfeifer’s method by replacing the least squares method with a normalized least squares method called adaptive line enhancement (ALE). The implementation of ALE required a priori knowledge of a number of parameters such as the delay factor and the adaptation parameter. In another study, iterative regression was also used by Brandtberg et al. (2003) to derive DEM in a forest area. 2. Morphological Methods: The idea of morphological methods is approximating the terrain surface using morphological operations such as opening. Compared with other methods, morphological methods are conceptually simple and can be easily implemented. When there are enough pulses reaching the ground, morphological opening with a small window size can effectively remove the surface objects and generate a surface approximating the ground. However, when there are not many pulses hitting the ground, such as the places where buildings are located, the window size for morphological opening has to be large to remove the objects. The problem of using a morphological opening with larger window sizes is that it will produce a surface with more protruded terrain features flattened. Therefore, how to keep the terrain features unchanged while using large window sizes for opening is the biggest challenge. PHOTOGRAMMETRIC ENGINEER ING & REMOTE SENS ING Feb r ua r y 2007 175 Qi Chen is with the Center for the Assessment and Monitoring of Forest and Environmental Resources CAMFER), University of California at Berkeley, Berkeley, CA 94720 ([email protected]). Peng Gong is with the Center for the Assessment and Monitoring of Forest and Environmental Resources CAMFER), University of California at Berkeley, Berkeley, CA 94720 and also with the State Key Lab of Remote Sensing Science (jointly sponsored by Institute of Remote Sensing Applications, Chinese Academy of Sciences, and Beijing Normal University), Postal Box 9718, Beijing, P.R. China 100101 ([email protected]). Dennis Baldocchi is with the Department of Environmental Science, Policy, and Management, University of California at Berkeley, Berkeley, CA 94720 ([email protected]). Gengxin Xie is with the School of Environmental Science & Engineering, Hunan University, Changsha, Hunan, 410082, China. Photogrammetric Engineering & Remote Sensing Vol. 73, No. 2, February 2007, pp. 175–185. 0099-1112/07/7302–0175/$3.00/0 © 2007 American Society for Photogrammetry and Remote Sensing 05-047 1/11/06 3:06 AM Page 175