Experiments on Generation of 3D Virtual Geographic Environment Based on Laser Scanning Technique Jie Du 1, Fumio Yamazaki 2 Xiaoyong Chen 3 Apisit Eiumnoh 4, Michiro Kusanagi 3, R.P. Shrestha 4 1 School of Civil Engineering, Asian Institute of Technology, Thailand 2 School of Civil Engineering, Asian Institute of Technology, Thailand Institute of Industrial Science, University of Tokyo, Japan 3 School of Advanced Technologies, Asian Institute of Technology, Thailand 4 School of School of Environment, Resources and Development, Asian Institute of Technology, Thailand Tel: +66-2-524-5905, Fax +66-2+524-6059/5905 miyakojie@yahoo.com Abstract Airborne laser scanner is an integrated system of GPS, INS and laser scanner. It has become an accurate, fast, and versatile measurement technique that can complement or partly replace other existing acquisition technologies and has been used in more diverse applications that concern large-scale and precise topographic Digital Elevation Models (DEMs) and Digital Surface Models (DSMs). Up to now, development of algorithm and methods for acquiring 3D spatial data, interpretation and modeling of this kind data for property application are main research topics. This paper demonstrates an experiment on generation of 3D virtual geographic environment by using experimental flight laser scanning data, based on a set of algorithms and methods that were developed for automatically interpret range images for extracting geo-spatial features to reconstruct geo-objects. The set of algorithms and methods for interpretation and modeling of laser scanner data include the followings: Triangulated Irregular Network (TIN) based range image interpolation; Mathematical Morphology (MM) based range image filtering, feature extraction and range image segmentation, feature generalization and optimization; 3D objects reconstruction and modeling; Computer Graphics (CG) based visualization and animation of geographic virtual reality environment. The results of the experiment of this research demonstrate the operational utilization of the suggested approach and its applicability to various fields of study, such as urban disaster management. Keywords: Laser scanning data, Virtual Geographic Environment (VGE), urban disaster. 1. Introduction Recently, the need to describe larger scale 3D spatial data has been continuously increasing. The reason is that, compared with two-dimensional data, threedimensional data can describe details of objects on the ground and enable conducting more accurate and detailed simulation or analysis on proper areas. Airborne laser scanning represents a new and independent technology to obtain highly automated generation of digital terrain models (DTMs) and digital surface models (DSMs). Nowadays, most technical hardware difficulties and system integration problems have already been solved. The remaining problem is the development of algorithms for interpretation and modeling of laser scanner data (Axelsson, 1999). By integrating laser scanner data with other existing 2D/3D data, reliability of high-resolution three-dimensional data generated will be more reliable. This study describes the system configuration and methodology of acquisition, processing and modeling 3D spatial data for generation of 3D virtual geographic environment, based on experimental flight laser-scanning data. This research proposes a new approach to develop a set of algorithms and methods for processing, modeling the 3D spatial data based on the laser range data in order to generate 3D virtual environment. This new approach for processing and modeling the 3D spatial data is a mathematic morphology (MM) based multi-resolution method. Due to the data availability, Kyoto Station area of Kyoto city in Japan, was chosen as the study area 2. Statement of the problem and methdology of reserach Today, the laser scanning system is offered by a very limited number of companies. Similarly, most technical hardware difficulties and system integration problems have already been solved. However, airborne laser scanner data have not been applied widely because the processing of this kind of data is still under research. Since the processing of airborne laser scanner data alone cannot provide a good result, it needs to be integrated with other existing 2D data sources. At this moment, 102 Du, J. and Yamazaki, F.
problems of various kinds are under research in order to improve accuracy of the result and reduce total working time. Up to now, development of algorithm and methods for acquiring 3D data, interpretation and modeling of laser data for property application are main research topics. This research aims to develop the algorithms and methods to automatically/semiautomatically interpret range images of extracted geospatial features and reconstruct geo-objects to generate 3D virtual environment based on laser scanner data. These algorithms mainly include: 1. Triangulated Irregular Network (TIN) based range image interpolation and Pre-processing; 2. Mathematical Morphology (MM) based range image filtering, features extraction and range image segmentation, feature generalization and optimization, 3D objects reconstruction and modeling; 3. Computer Graphic (CG) based visualization and generation of virtual environment. Laser range images can be kind of high-resolution digital terrain models (DTMs) or digital surface models (DSMs). Reconstruction of 3D objects from laser range images can simply be considered as the 3D raster-vector conversions from a noisy DTM to parametric CAD formatted 3D vector data. Due to the limited resolution and noise, the existing low-level processing methods for reconstruction of 3D objects from laser range images generally cannot be used for real application needs. City and regional planning maps and 3D GIS data, a set of well-organized high level data source interpreted and generalized by human operators, are useful knowledge for 3D object reconstruction and recognition. 3. Case Study 3.1 The study area The ALS (Airborne laser scanner) data employed in this research was acquired over Kyoto railway station area of Japan, provided by Asian Air Survey Co., Ltd. The basic specifications and data parameters are given in Table 1. Aerial photographs and city maps are used as additional information in feature extraction and others related processing. The city map of Kyoto Station area, Kyoto city planning GIS database were used in this research. This database was supplied by Kyoto City Urban Planning Bureau. Table 1 Specification of ALS employed in this study Provided by Asian Air Survey Co., Ltd Specification of ALS employed in this study Flying height 200 400 m above ground Platform Helicopter Resolution 50 cm on the ground Pulse Rate (Hz) 20 khz Scan Rate (khz) 25 Hz Scan angle ±30 degrees Scan Swath Width 0.68Xh m Cross Track Spacing 1.93 m Along Track Spacing 2.83 m X, Y Positional Accuracy 1/2000 X h Z Positional Accuracy 0.15 m RMSE absolute Beam divergence (mrad) 1.5 3.2 Proceduce 3.2.1 TIN based pre-processing As the first step, the raw data recorded by the laser ranger need to be translated to a file format that SoftImage 3-D can read. Generally, there are two ways to translate the data. The first way is to generate a SoftImage 3-D graphic file directly. The second way is to translate a DXF file (generated by MicroStation or AUTO-CAD) into a SoftImage 3-D file. Both of these methods for inputting laser range data are useful for a different laser range hardware. Have tried these two methods for imputing laser range data and they are useful for different laser range hardware (such as airborne or ground-type machines). Another important issue is the conversion of raw data from their local coordinate systems to global ones. The processing can be realized by making C program for 3-D coordinate transformation based on a set of provided parameters. Figure 1 shows the result after the processing of TIN (Triangulated Irregular Network) based range data preprocessing and range interpolation. Figure 1 result after the processing Du, J. and Yamazaki, F. 103
3.2.2 MM based processing Mathematical Morphology (MM) based approaches are used (Chen. and Ikeda, 1994) for the purposes of range image filtering and related object segmentation and semi-automated feature extraction. MM operators (such as dilation, erosion, opening, closing, hit or miss, thinning) can be described as a combination of shift and logic operations. Shifting operations are controlled by the given Structuring Elements (SEs) whose size, shape, and orientation can be changed by different applications. Different MM operators organize different logic operations for different purposes. Figure 2 procedure of MM based processing MM filtering and 3D object segmentation are one of the most successful tools used for MM applications. For range image processing, the opening filter was used to remove dirty voxels and small-connected volumes. The closing filter was also used to fill small holes within surfaces and to link short gaps among objects. Here, the key problem is how to select suitable SEs. Generally, before the real processing of whole large images several, typical, small testing areas can be processed to find the suitable parameters and optimal processing procedures. Also, simple knowledge bases can be generated based on these selected parameters and procedures that can be used for other range image processing. The basic idea of MM based object segmentation is: firstly, filtering all the parts smaller than the given SEs; then, segmenting the objects by logic difference operations between the original 3-D data set and the filtered 3-D data set. Since MM based filtering with a small SE in the first step of processing will damage the detail object features, a feature-recovering step should be added during the object segment procedure. When using the 2D map data for range image processing, a feature protected filtering based on the conditional masking volumes generated by the 2-D boundary lines of building or roads can also be realized. Figure 2 shows a simple procedure of MM filtering and object segmentation. A sample result of the laser range image of Kyoto railway station area by using MM based multiresolution processing algorithms and methods that were utilized in this research are shown in Figure 4. Figure 3 shows the result by using other methods while figure 5 shows a laser range image without further processing. The area marked by a circle in Figures 3, 4 and 5 is a tall building (Kyoto Tower). The actual shape of the building when viewed from its side is tapering. The base has a wider width compared with a smaller width of the top floor. After filtering and segmentation, the MMbased method produced better in terms of noise reduction. In Figures 4, the noise is filtered very efficiently. Circle layers of the building were realized. While the circle layers in Figure 5 is unclear. This filtering utilizes the minimum-threshold method. Most features were eliminated with this method. This means that even if the elimination of the noise is effective, other significant information of the object as well should not be lost. In this case, the relative tapering of the building as height increases was also eliminated. The cross-section profile (Figure 6) of an object was created to discuss the methods in detail at different range images as an example to analysis the profile graphs. This is for quantitatively analyzing the results of the different methods. The solid line represents filtering and segmentation process using the MM-based method, dashed line represents the cross-sectional image after minimum-threshold filtering and the dotted line represents a raw laser range image without any feature processing. The MM-based method clearly defines most of the features of the object compared to the raw range image and even after utilizing the minimum-threshold method. 3.2.3 CG based processing CG (Computer Graphic) based processing includes 3D visualization for generation of virtual geographic environment. To generate a complex 3-D object with surfaces, 2D image textures from air photos. By using the SoftImage 3-D system, 3-D spatial objects based on extracted spatial lines and surfaces can be generated. Figure 7 shows the procedure of visual texture mapping processing to generate 3D virtual geographic environments. 104 Du, J. and Yamazaki, F.
60.10 P ro file G ra p h Vertical exaggeration 1.5 X Elevation 28.34 38.34 48.34 0 10.00 20.00 30.00 40.00 50.00 Distance MM based method Without processing Other method 64.15 Figure 6 Profile graphs of different range images Figure 3 Result by using Minimum-threshold Filtering method Figure 4 Laser range image by using MM based multiresolution processing Figure 7 Flow from texture mapping to result 4. Testing system A system based on the algorithms and methods of the research processing was developed. This system consists of four main modules. The developed four main modules include display, processing modeling and texture. Figure 8 shows the interface of this system (main window). Figure 5 image of the range image without feature processing Du, J. and Yamazaki, F. 105
(a) (b) Figure 8 Interface of the developed processing system 5. Conclusions The study, experiments, and issues addressed in this paper were integrated to the following conclusions: 1. Laser scanning is a suitable technique to collect relief information to be used for 3D virtual environment generation. A promising development is the combination of an existing database with the automatically extracted features to facilitate the time-consuming task of manual methods. 2. A special system based on C++ and SoftImage 3D had been developed due to the unavailability of commercial software for this kind research. This system was designed based on the algorithm logic structure and other procedures. Though it is only a prototype, it still met the rigorous demand of the study, e.g. the system developed in this research can handle the enormous (laser random point) data. A special commercial software was needed for this kind of processing. 3. The developed algorithms, methods, and system were successfully applied to study area. The results of the study satisfied the research objectives. The procedures formulated for automatic interpretation and segmentation of laser range data were successfully applied in this study using MM based multi-resolution processing methods. The study and its applications have proven the feasibility and success of the developed methods and, it is concluded, the methods will be useful for visualization of 3D objects and landscapes. 4. The point density acquired by airborne laser scanner used in this research was unsatisfactory for some case in some degree. For instance, in the vegetation area, after interpolation of points to grid, it is difficult to distinguish points indicating the trees from those points referring to the ground. In addition, the shape of top parts of high objects (i.e. buildings) does not match the sharpness of the objects very well. 5. The methods achieved in this study have advantage in term of high automation and fast delivery times comparing the one attained by photogrammetric stereo compilation and other methods. 6. From the processing result, the algorithm and processing system in this research was developed in an open way. There are remaining several problems: The laser data used in this research was only a one-time flight data. Due to limited scanning direction and flying speed, some areas were unreachable by the laser beam. Random data of these areas are gapped, such as foot parts of high objects, and the side of objects surface that are opposite to the direction of laser beam emitting. Therefore, some site have no 3D data such as gap areas, break-points (lines problem) etc. 6. Recommendations Based on the research conducted, a number of recommendations for future research have been summarized. 1. As mention in the conclusion, the use of height data for 3D building reconstruction has been demonstrated. By the integration of ground plans and GIS data, detailed reconstructions of buildings can be obtained automatically even for laser data measured at relatively low point densities. Since the integration of laser data proved very successful, the use of this information is strongly recommended during the automatic generation of density area databases. E.g. building urban building inventory environment after natural disaster like flooding, earthquake, etc.. 2. Automatic procedures for interpretation and segmentation of laser range data can be successful for the applications by using MM 106 Du, J. and Yamazaki, F.
based processing methods. For a more general approach, other sources of information should be used in order to raise the success rate of the procedure. Such sources of information are reflectance data or multiple echoes from the ALS system, images, existing 2D GIS databases, land use maps, etc. Therefore, laser data are just random point data, these are more suitable for mapping purposes and needs to be merged very closely with the image information. This can be achieved either by using oriented images from photogrammetric image blocks, supplied directly from the same platform as the raw Laser data or from the laser raw data itself in the form of a reflectance image. 3. The study integrated existing maps, GIS data and aerial photographs with processed laser scanning data. Further research is therefore recommended to concentrate on the design of an improved procedure on attaining higher accuracy and automation. These can be any of the following points: ٠ Improving the accuracy of object structures by using 3D data from multisources (Mobile mapping, laser range data from ground stations or multi-times scanning data); ٠ Automated 3D feature extraction based on knowledge bases and existing GIS; ٠ Improving the accuracy of texture images. ٠ Generation of orthophotos for accurately texture mapping Other possible approaches of integrating multisensor data and other various applications 4. Based on laser scanning techniques to generation of 3D geographic environment have been used in some developed countries in urban systems for infrastructure design, urban planning and management. This technique can be an expecting/ effective way for acquisition 3D spatial data for simulating and monitoring disaster situation for demonstration and assessment of damage. It is with advantages that are: ٠ Real-time mapping and simulation ٠ Easy acquisition ٠ High accuracy ٠ High automation and fast delivery times It can be become an effective tool for rapid and real time acquisition of demanded data and construct the database for disaster evaluation as well as producing 3D view to show the damage spatial distribution visually. Therefore, it can be applied to evaluate the urban risk of the area in the developing country that urban infrastructures data are still not sufficient, like most GMS country. Acknowledgement The experiments and principle was based on the doctoral dissertation research: Generation of 3D Virtual Geographic Environment. Gratitude to whom that have provided various valuable comments and guidance throughout the study and research. Reference Ackermann, F., 1999. Airborne laser scanning present status and future expectations. ISPRS Journal of Photogrammetry & Remote Sensing 54, 64-67. Chen, X., Ikeda, K., 1994. Three Dimensional Modelling of GISBased on Delaunay Tetrahedral Tessellations. Proc. ISPRS, Comm. III Symposium, Munich, Germany. In: Int l Arch.Photog. and Rem. Sens., Vol. XXX, Part B3 r 1, pp. 124 131. Foerstner, W.; Guelch, E. 1999. Automatic Orientation and Recognition in Highly Structured Scenes. ISPRS Journal of Photogrammetry and Remote Sensing 54, 23-34. Faugeras, O.,1993. Three-Dimensional Computer Vision: A Geometric Viewpoint. The MIT Press, 663 pages Axelsson, P., 1998. Integrated sensors for improved 3D interpretation. International Archives of Photogrammetry and Remote Sensing v 32, 27 34, Part 4. Du, J. and Yamazaki, F. 107