|
1.INTRODUCTIONRecently, the one map policy in Indonesia began to be encouraged for large-scale maps up to 1/10.000 - 1/5.000. Hence all efforts to encourage the acceleration of large-scale map production are always needed. Using the Very High-Resolution Satellite (VHRS) imagery for mapping can be a good choice, due to its wide-area coverage. There are many kinds of researches that have used of VHRS imagery technique for producing a base map (see [1], [2], [3]). In general, The Base Map use DTM (Digital Terrain Model) as elevation information. Unfortunately, producing terrain point cloud from airborne or spaceborne with optical sensor camera is quite the challenge. The elevation point model extraction from spaceborne or aerial platform imagery only produces DSM (Digital Surface Model) instead of DTM. However, working with an aerial photo and VHRS image can produce a high spatial resolution that meets the need for large scale mapping. This study uses Worldview-3 stereo ready images that have a 30cm x 30cm spatial resolution. While the aerial photo uses 10cm Ground Sampling Distance (GSD). Furthermore, the DSM filtering process is one of the procedures that can be done for classifying terrain and non-terrain point cloud. The goal of this study is to analyze the suitability of the filtering methods towards the generation of Digital Terrain Model (DTM) from Digital Surface Model (DSM) extraction. Some DSM filtering algorithms to classify between terrain and non-terrain point cloud has tried. There are three kind algorithm, which are: (1) Elevation Threshold with Expand Windows (ETEW), (2) 2D Morphological Square (Morph 2D Filter), and (3) Adaptive TIN (A-Tin) (see [4], [5], [6], [7]). The ALDPAT (Airborne LIDAR Data Processing and Analysis Tools) v.1.0 software was used to implement those algorithms. Although the ALDPAT program is designed for handling LIDAR data, it can be used for general point cloud data also. These filtering algorithms are implemented for residential and vegetation areas. The residential area mostly covered by houses objects. While vegetation area rather open area. The evaluation test has been made by visual interpretation to see the point remain. A good filtering result can remove a non-terrain point as much as possible. This visual interpretation can be done by using correspondence its orthoimage and orthophoto. Furthermore, an interpolation algorithm has to be done to fill the area missing point. 2.IMAGE DATASHEETThis study uses two kinds of landscape area which are (1) Mountainous area at The Tasikmalaya District and (2) Sub-Urban area at Probolinggo District. Table 1 shows the sample image of Orthoimage from Worldview-3 and Orthophoto from aerial photo. The DSM data from Worldview-3 has derived by stereo parallax with area-based image matching from stereo epipolar image. While the DSM from the aerial photo has derived by the parallax algorithm also. Table 1.Image datasheet. 3.METHODOLOGYThe methodology is arranged to fulfil the research objectives effectively (see Figure 1). The most important steps are described as follows: This research performs several phases of the procedure. First, acquisition of the DSM Point Cloud data from the WorldView-3 and DSM Point Cloud data from aerial photo data. Data is done filtering process using ALDPAT software in generating DTM. The DSM filtering process for both data uses the Method 3 algorithm which is ETEW, ATIN, and Morph 2D so that obtained 3 outputs from the DSM filtering process of the ALDPAT software. Once the cloud point is obtained from the DSM filtering process, the DTM resulting from the use of 3 algorithm methods, performed a visual evaluation of the data. As for the concept explanation of the 3 algorithm methods used as follows:
(1) The three option Filtering algorithm are as follow: ETEW: Elevation Threshold with Expand Windows (see [5], [7]) Width: cell width Height: cell height Max Z: maximum elevation allowed for all ground points Min Z: minimum elevation allowed for all ground points Slope: slope factor in determining the threshold value of elevation differences Loop Times: maximum number of iteration performed by the filter ATin: Adaptive TIN (see [6], [7]) Cell Size: cell size of a grid to parse the raw data. Z difference: the threshold used to compare the distance from a point to a point beneath the TIN surface. Angle Threshold: the default value is 0 where no angle threshold is used. Init TriGrid Size: the initial size of a grid used to select seed points for the ground data set. A point with a minimum elevation within a grid cell is selected as a seed point. Tile X Width: the width of a rectangle for applying ATIN filter. Delauney triangulation of LIDAR points for a large area is time consuming. Tile Y Height: the height of rectangle for applying ATIN filter. Tile Buffer: buffer size for a rectangle. Morph 2D Filter: 2D Morphological Square (see [4], [7]) Cell size: cell size. Slope: slope value used to determine the elevation difference threshold. Init Threshold: the initial elevation difference threshold which approximates the error of LIDAR measurements (0.15-0.3 m). Max Threshold: the maximum elevation difference is an optional threshold which is set to a fixed height (e.g., the lowest building height) to ensure that building complexes in urban areas are removed Window Base: base to determine the window size. Power Increment: increment for computing a window series. Window Series Length: length of a window series. Init Radius: initial search radius for nearest neighbor interpolation Wind Series: window size series. Window size is computed by a power function: (window base)(window step). window size and slope value. (2) The visual interpretation has done by using its correspondent orthophoto. A terrain point cloud should be remain, while the non-terrain point should be removed. Plotting remain point cloud at its orthophoto can be used to evaluate the filtering quality. 4.RESULT AND DISCUSSION4.1DSM FilteringPoint cloud data from DSM filtering results with ALDPAT program can be seen in Table 2. and Table. 3The filtering results from the three algorithms produce different point removing in the same area. Experiments in residential and vegetation areas provide different responses by using difference filtering algorithms. Table 2.Result from DSM Filtering in Residential Area Table 3.Result from DSM Filtering in Vegetation Area Table 2 explains the visualization of filtering results by using 3 algorithm methods, including ETEW, ATIN, Morph 2D method in the residential area. The ETEW method for residential areas, there are still some clouds on the roof but can be removed, on this method it still leaves a lot of dots. The ATIN method for the residential area is a method with the best results that point on the roof can be removed. The 2D Morph method for residential areas there are still some cloud spots on the roof but lacking many still leave some point. Table 3 explains the visualization of filtering results by using 3 algorithm methods, including ETEW, ATIN, Morph 2D method in the vegetation area. On the ETEW method for the vegetation area for the tree object can be deleted still leave some dots. On the ATIN method for vegetation is a method with the best results with dots on the tree object can be removed. In the method of the 2D Morph for vegetation, there are still some points on tree objects but less numerous. The comparison between three algorithm shows that Adaptive TIN give the best result. The ATin method provides the best results because the number of non-terrain point clouds can be removed rather dominant. ATin algorithm also works both in residential and vegetation area. Adaptive TIN algorithm has the ability to handle surfaces with discontinuities deviation. Unfortunately, there are still some non-terrain point exist. This is because the parameter setting cannot working for removing all kind of objects in the earth surface. In this case, the process should be repeated until all the non-terrain point has removed. There is no guarantee that parameter algorithm can working for all landscape area also. 4.2Final DTMProcessing DSM filtering produce bare-earth point that close to terrain point. While the non-terrain point removed. This condition make missing point in some areas. The missing point should be filled by interpolation. Figure 2 and 3 shows the comparison between DSM and DTM derived from filtering procedures in vegetation areas. It can be shown that there is still some non-terrain point at dense vegetation and residential area. This non-terrain point makes contour like bull eyes shape. The non-terrain point remaining can be detected as spike or pit in elevation value. It is need to repeat filtering procedures for removing remaining non-terrain point. 5.CONCLUSIONThis paper shows some algorithm for DSM filtering to remove a non-terrain point cloud. The use of different algorithms produces different remaining terrain points. The Adaptive TIN algorithm give the best result for DSM filtering procedure followed by ETEW and Morph 2D algorithm. The filtering algorithm with ATin and ETEW can working in residential and vegetation at any landscape area. While Morph 2D algorithm can be used in flat area. There are still some non-terrain point remaining after DSM filtering, because the setting parameter algorithm cannot working for varies object at the surface of the earth. In this case, the iteratively filtering process should be made. Furthermore, an interpolation process should be made for filling the missing non-terrain point. REFERENCESCheng, Ph., Chaapel, C.,
“Using WorldView-1 stereo data with or without ground control points,”
GEOinformatics, 11
(7), 34
–39
(2008). Google Scholar
Hu F., X.M. Gao, G.Y.Li M.Li,
“DEM Extraction From Worldview-3 Stereo-Images and Accuracy Evaluation, The International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, Volume XLI-B1, 2016 XXIII ISPRS Congress,”
327
–332 Prague, Czech Republicp2016). Google Scholar
Capaldo, P.,Crespi, M., Fratarcangeli, F., Nascetti, A.,Pieralice, F.,
“DSM generation from high resolution imagery: applications with WorldView-1 and GeoEye-1.,”
European Journal of Remote Sensing, 44
(1), 41
–53
(2012). Google Scholar
Zhang, K., Chen, S., Whitman, D., Shyu, M., Yan, J. and Zhang, C.,
“A progressive morphological filter for removing non-ground measurements from airborne LIDAR data,”
IEEE Transactions on Geoscience and Remote Sensing,, 41 872
–882
(2003). https://doi.org/10.1109/TGRS.2003.810682 Google Scholar
Zhang, K. and Whitman, D.,
“Comparison of three algorithms for filtering airborne LIDAR data,”
Photogrammetric Engineering and Remote Sensing,, 71 313
–324
(2005). https://doi.org/10.14358/PERS.71.3.313 Google Scholar
Axelsson, P.,
“DEM generation from laser scanner data using adaptive tin models,”
International Archives of Photogrammetry and Remote Sensing, XXXIII, 85
–92
(2000). Google Scholar
Sithole, G. and Vosselman, G.,
“Experimental comparison of filter algorithms for bare-Earth extraction from airborne laser scanning point clouds,”
ISPRS Journal of Photogrammetry and Remote Sensing,, 59 85
–101
(2004). https://doi.org/10.1016/j.isprsjprs.2004.05.004 Google Scholar
Axelsson, P.,
“DEM Generation From Laser Scanner Data Using Adaptive TIN Model,”
International Archives of Photogrammetry and Remote Sensing, 33 Amsterdam. (2000). Google Scholar
Sulaiman N S, Majid Z, and Setan H.,
“DTM Generation From Lidar Data By Using Different Filters In Open – Source Software,”
Geoinformation Science Journal, 10
(2), 89
–109
(2010). Google Scholar
|