Surveying and Mapping Caves by Using 3D Digital Technologies

Transcription

1 Surveying and Mapping Caves by Using 3D Digital Technologies Wei Ma and Hongbin Zha National Laboratory on Machine Perception, Peking University, , China Abstract. In this paper, we propose an original three-dimensional (3D) computer-aided approach for surveying and mapping caves to get line drawings. By introducing 3D digital technologies into the line drawing, our approach has advantages in improving the drawing accuracy while cutting a lot of time costs. The 3D digital technologies are used to reconstruct accurate 3D models of real scenes. Then, the lines on the model surfaces are extracted by analyzing the model geometry. To demonstrate the advantages, we compare our approach with the traditional method usually used by archaeologists. Results show that our approach is much more convenient and accurate than the previous method. 1 Introduction Line drawings (plan, elevation and profile) are a necessary part of archaeological reports for caves. To get line drawings, archaeological workers usually use a traditional manual method. For example, to draw a profile of a cave, the workers do the following steps: 1. decide two standard baselines perpendicular with each other by poles and ropes. One is parallel to the horizon along a profile plane, while the other is vertical to the horizon; 2. mark important features manually on the walls of the cave; 3. measure each feature by using rulers or spreading grids based on the coordinate system determined by the two baselines. To facilitate surveying, additional lines parallel to the baselines are added, each with a soft ruler attached, as shown in Fig. 1(a); 4. draw the features in a millimeter paper and connect them to produce a smooth drawing. The traditional procedure above is labor intensive and time consuming. Furthermore, most of the sculptures in grotto sites are weathered, eroded and oxygenated to be seriously mottled [1]. It is difficult to recognize their original carved appearances as shown in both pictures of Fig. 1. In this case, manual surveying is subjective and inaccurate. Since the traditional methods are subjective and time consuming, it is necessary to find a new method to simplify and automate the procedure. Therefore, the School of Archaeology and Museology, the National Laboratory on Machine H. Zha et al. (Eds.): VSMM 2006, LNCS 4270, pp , c Springer-Verlag Berlin Heidelberg 2006

2 Surveying and Mapping Caves by Using 3D Digital Technologies 369 Fig. 1. (a) A traditional working site; (b) a scanning working site Perception at Peking University and the Longmen Grottoes Research Academy make an experiment on line drawings aided by three dimensional (3D) digital technologies in the Leigutai area of Longmen Grottoes. Parallel projection maps without any colors and perspective distortions are provided by our laboratory. Archaeological workers in Longmen Grottoes are responsible for the line drawing from the maps. The paper describes the whole process in detail. Fig. 2. Pipeline of our procedure Our working pipeline is shown in Fig. 2. In the first step, we scan an object from multiple planned viewpoints by using laser scanners. Then we pre-process each view-scan, register all the scans into a single coordinate system, merge them to be a whole model and fill holes on the surface of the model. During post-processing, we render the model in proper lighting conditions and clip it into different parts for its plan, elevation and profile parallel projection maps under the instruction of archaeological workers in Longmen Grottoes. At last, the workers spread each map under a millimeter paper, and draw lines on the basis of the maps.

3 370 W. Ma and H. Zha In Section 2, we describe how to model a real object, which includes two steps: data acquisition and data processing. In Section 3 we show how to create line drawings from 3D models. Experiment results are given in Section 4. Section 5 presents conclusions. 2 Model Reconstruction 2.1 Data Acquisition To create an accurate 3D model, we scan the surface of a real object from multiple viewpoints using laser scanners, such as Cyrax 2500 and VIVID910. The main parameters of the two scanners are listed in Table 1. For efficiency, we use Cyrax 2500 (Fig. 3(a)) with lower precision to scan large caves, and VIVID910 (Fig. 3(b)) with higher precision for fine details. We change viewpoints by moving scanners. Before scanning the object, we first decide the positions of all needed viewpoints around the object. This procedure is called view planning. The goal of the view planning for the object is to cover its whole surface with as few as possible view-scans, while keeping neighbour scans partially overlapped as illustrated in Fig. 3(c). For surfaces inaccessible from the ground, special shelves for the scanners are built. Fig. 1(c) shows a planned position for scanning a high and concave part in the south cave of the Leigutai area. Table 1. Parameters of Cyrax 2500 and VIVID910 Scanners Scan range Field of view Precision Cyrax 2500 < 150m mm VIVID910 < 2.5m mm Fig. 3. (a) Cyrax 2500; (b) VIVID910; (c) a two-viewpoint illustration

4 Surveying and Mapping Caves by Using 3D Digital Technologies Data Processing We acquire the data of an object view by view as described in section 2.1. Some obtained view-scans may contain noisy parts because of self-occlusions, etc. During pre-processing, we delete the obvious noisy parts. On the other hand, processing large data with high precision is difficult even with graphic workstations. We decimate the original scanning data also in the pre-processing step. Fig. 4. The Bodhisattva on the main wall of the middle cave (a) after registration; (b) after merging Scanning a large cave using Cyrax2500 or a fine statue using VIVID910 often produces tens of view-scans. These scans are independent and each has its own coordinate system. We use the IMAlign module in Polyworks, a software produced by InnovMetric Software Inc., to register all the scans and put them into a common coordinate system [2]. The main steps are: 1. import two view-scans, and choose their corresponding feature points respectively; 2. regarding the first scan s coordinate system as the world one, compute the second one s transformation matrix relative to it based on the matched points and transfer the second scan into the world coordinate system; 3. obtain an optimal alignment using the ICP (Iterative Closest Points) algorithm by minimizing the 3D distances between surface overlaps. Taking the two scans as a whole, we then import a third one and follow the three steps. The same operations are carried out to the other scans, until all of

5 372 W. Ma and H. Zha them are in the same coordinate system, as shown in Fig. 4(a) with different colors denoting different view-scans. After the registration, we use the IMMerge module in Polyworks to merge the different view-scans into a monolayer triangle mesh model as shown in Fig. 4(b). Due to occlusions, some surfaces are impossible to scan, resulting in holes on the surfaces of 3D models. Holes on simple and smooth surfaces can be repaired automatically with allowable errors. We fill these holes with triangles and subdivide the triangles into finer ones with uniform surface resolutions by using the IMEdit module of Polyworks. Yet, holes lying on complex surfaces are difficult to be filled automatically and accurately with existing hole filling algorithms [3, 4]. Repairing them manually is inaccurate and time consuming. Instead of filling these holes here, we keep them until the line drawing step, where archaeological workers add feature points manually on the missing parts of the line drawings by measuring their corresponding points on the real objects. 3 Line Drawing After getting an object s 3D model, we render it under proper lighting conditions and clip it to get the model s plan, elevation and profile parallel projection maps. Then, archaeological workers spread each map under a millimeter paper, and finish line drawing by referring to the maps. Fig. 5 shows a plan parallel map of the north cave at 1.3 meters high and its line drawing. Fig. 5. A plan of the north cave at 1.3 meters high. (a) Parallel projection map; (b) line drawing. In the research of 3D digital geometry processing, feature lines on model surfaces can be extracted automatically based on the surface variations [5, 6]. Decarlo [5] detects lines by setting a threshold for the inter-angles between the

6 Surveying and Mapping Caves by Using 3D Digital Technologies 373 normal direction of each point on the model surface and a current view direction. Points with angles larger than the threshold are defined as feature points. Yoshizawa [6] introduces crest lines on the model surface as features. The crest lines detection is based on estimating curvature tensors and curvature derivatives via local polynomial fitting. Fig. 6(a) shows a line drawing extracted by us using Decarlo s method. However,usually the lines extracted automatically do not satisfy the requirements of archaeological reports. An archaeological line drawing (as shown in Fig. 6(b)) needs to be partially emphasized. Realizing these requirements rely on archaeologists subjective judgements and artistic knowledge. Fig. 6. The Bodhisattva on the main wall of the middle cave. (a) Lines extraced automatically; (b) lines drawn manually. Although an automatically extracted line drawing does not satisfy archaeological requirements, it includes all needed features and seems closer to an archaeological one than its corresponding projection map. In the future, instead of the line drawing by projection maps, we hope to provide archaeologists an ideal platform which can extract accurate and necessary feature lines from a 3D model automatically and also includes an interactive interface for users to edit the lines. With this platform, people can get archaeological line drawings conveniently. 4 Results Fig. 7(a) and (c) are the parallel projection maps of the Buddha statue on the main wall of the north cave and the outside elevation of the Leigutai area

7 374 W. Ma and H. Zha Fig. 7. The Buddha in the north cave. (a) Parallel projection map; (b) line drawing. The outside elevation, (c) Parallel projection map; (d) line drawing. respectively. Fig. 7(b) and (d) are their corresponding line drawings. Table 2 presents the time costs for the traditional method and our method on surveying and mapping three different objects: the Bodhisattva on the main wall of the middle cave (Fig. 4, Fig. 6), the Buddha statue on the main wall of the north cave (Fig. 7(a) and (b)), and the outside elevation (Fig. 7(c) and (d)) in the Leigutai area. Comparing with the traditional method, our method saves 52 hours for the Bodhisattva in the middle cave, 324 hours for the Buddha in the north cave, and 448 hours for the outside elevation. The larger and the more seriously eroded

8 Surveying and Mapping Caves by Using 3D Digital Technologies 375 Table 2. Comparison between the computer aided method and the traditional method in time costs Objects Size(m 2 ): width height View number computer aided(h): modeling+line drawing Traditional method(h) Saved time(h) Bodhisattva in the middle cave (VIVID) Buddha in the (Cyrax) north cave Outside elevation (Cyrax) (VIVID) theobject,themorethetimesaved.wepresentonelinedrawingforeachofthe three objects above. Generally, all kinds of drawings (plan, elevation and profile) for an object can be produced quickly once its 3D model has been reconstructed as described in Section 3. Such a batch production ability for one object is a new feature as compared with the traditional method. Fig. 8. A Warrior in Liutian cave. (a) Photo; (b) parallel projection map. The Budda on the main wall of the north cave. (c) Photo; (d) parallel projection map. In accuracy, we make no quantitative comparison between the traditional method and our method for now. However, the traditional method suffers errors due to manual measuring, perspective distortions and the disturbances of mottled colors. In contrast, the measuring errors of our laser scanners are ralatively small and parallel projection could be easily realized in computers. In addition, our method is based on the geometric shapes of 3D models rather than any color information. Fig.8 (a) and (c) are photos of a Warrior statue in Liutian cave and the Buddha statue on the main wall of the north cave respectively. Fig.8 (b) and

9 376 W. Ma and H. Zha (d) are their corresponding projection maps used for the line drawing. Archaeologists seeing the maps were strongly impressed because they never felt before that these statues were carved so elaborately. These examples show that the computer aided line drawing without any color disturbances are certainly more accurate than the traditional manual method. In some other applications, such as building a digital museum [7], we generally take photos as textures for reality. However, there is no necessity to do the work for the line drawing purpose. 5 Conclusions In this paper, we integrate 3D digital technologies into surveying and mapping caves and describe the whole process. By comparing with the traditional manual method, we show that the new approach is much more efficient and accurate. However, there still exist two issues unresolved. One is the lack of an accurate and automatic hole-filling algorithm as described in Section 2.2. The second problem is the limitation of computer hardware for processing high precision scanning data so that we have to decimate the original data for practice. If the two problems were resolved, the quality of line drawings would be improved largely. Acknowledgment The pilot project is supported by the School of Archaeology and Museology, the National Laboratory on Machine Perception at Peking University and the Longmen Grottoes Research Academy. We would like to thank Prof. Chongfeng Li, one of the main organizers, for his direction on the experiment, and Danhong Huang, Tao Luo, Xin Li, Tao Wei etc for their efforts in the data scanning and processing. References 1. Yan, S., Fang, Y., Sun, B., Gao, H.: Influence of water permeation and analysis of treatment for the Longmen Grottoes. Geoscience 19 (2005) InnovMetric Software Inc.: Polyworks user s guide (2001) 3. Davis, J., Marschner, S., Garr, M., Levoy, M.: Filling holes in complex surfaces using volumetric diffusion. The First International Symposium on 3D Data Processing Visualization and Transmission. (2002) Jun, Y.: A piecewise hole filling algorithm in reverse engineering. Computer-Aided Design 37 (2005) DeCarlo, D., Finkelstein, A., Rusinkiewicz, S., Santella, A.: Suggestive contours for conveying shapes. ACM Transactions on Graphics. 22 (2003) Yoshizawa, S., Belyaev, A., Seidel, H.: Fast and robust detection of crest lines on meshes. ACM Symposium on Solid and Physical Modeling. (2005) Li, X., Feng, J., Zha, H.: 3D modelling of geological fossils and ores by combining high-resolution textures with 3D scanning data. The ninth Intenational conference on Virtual Systems and Multimedia. (2003)