Study site
Zhanyuan, one of the four famous gardens in the Jiangnan area of China, is located in Nanjing. It was built during the Jiajing period of the Ming Dynasty, and its spatial layout and gardening techniques are typical of the private gardens in Jiangnan. Zhanyuan has Jingmiao hall in the center with a long corridor, and the layout is also equipped with pavilions, a windowed veranda, waterside pavilions, and other garden buildings to create a rich and varied garden space. There are two rockeries in the south and north of the courtyard, which are located on the north and south sides of Jingmiao hall. The southern rockery was built by Liu Dunzhen, a famous Chinese architect [33].
The overall shape of the mountain surrounds the lower front portion of the rockery and the higher backend as well. The peaks and mountains make up dispersed front and back layers with stone walls that are vertically divided by grooves. Thus, the limestone texture in other natural mountain walls is consistent. According to the principle dome structure, a cave is built on the stone wall, and an artificial waterfall is constructed alongside plants, making it appear like a valley in the mountains (Fig. 1).
Target-free 3D scanning method
In an early stage, the team performed a 3D scan of Suiyuan and Huanxiu Shanzhuang. By setting up the station, they were able to acquire field data from the 3D laser scanner, the Leica ScanStation C10. According to different garden-building elements, sweeping sites were set up for the whole garden. To splice point cloud data at a later stage, the scanning targets should be established as the site is being set up in the early stage in order to register the point cloud. During the scanning process of Suiyuan, thirteen scanning sites and thirteen scanning targets were set up. In Huanxiu Shanzhuang, however, eighty-nine scanning sites and sixty-one scanning targets were established, including twenty-two stations and ten targets in the main rockery area (Fig. 2).
During the scan of the entire Huanxiu Shanzhuang site, the target and site could not be set in the deep cave of the rockery and the narrow spaces, such as the stone joint, because of the site’s space limited for setting up the scanning target. The registration accuracy of the point cloud is affected by the number of targets. In the later stage, when using Cyclone (software for processing point cloud data) to match the point cloud on time, there were some problems in the rockery of Huanxiu Shanzhuang, such as the staggered stone area, leaks in the splicing process, and coordination confusion (Fig. 3). In the later stage, when the point cloud model was refined, manual registration was needed in areas that had fewer scanning targets (e.g., mountain caves), which resulted in a large error. Therefore, from the researcher’s perspective, it is feasible to use the 3D laser scanner with the scanning target type data acquisition method for researching a park’s landscape elements. However, for a rockery with a single garden element, the accuracy of digital rockery research is limited by an early field operation error.
To avoid the problems caused by Huanxiu Shanzhuang scanning and further improve the scanning accuracy, this experiment adopts a target-free 3D scanning method to avoid the problem of limited data accuracy caused by a target that cannot be placed. The object of this data collection is the South Rockery of Zhanyuan, which remains a complex space. Since the surrounding plants provide ample cover with dramatically changing heights, most areas are difficult to walk through. Meanwhile, because of the demand of the Chinese classical garden’s construction method, it is closely connected with the surrounding environment, and a large number of rockery components are shielded from each other. This environment cannot use a conventional data acquisition method, such as UAV aerial photography or photogrammetry. Although the stand-alone 3D laser scanner can ensure the data’s high accuracy, the complexity of the rockery and occlusion between the landscapes render many areas of the mountains difficult to either stand or scan. Moreover, the rockery is primarily composed of Taihu stone (a type of stone from Lake Tai, in the western part of Suzhou, China, which has a variety of forms, many holes and textures on its surface, and often appears in classical gardens), and so the surface has caves and an extremely complex texture. Relying only on the terrestrial 3D laser scanner means that there will be similar problems with surveying and mapping Huanxiu Shanzhuang and Suiyuan.
To avoid these problems and collect sufficiently accurate rockery data, this study attempts to use a hand-held 3D laser scanner for mapping. Compared to the stand-alone scanner, the hand-held scanner is highly flexible, and it can operate in local fine areas without setting up a station. Although it has a high operability capacity for caves, cliffs, areas with complex structures, and narrow spaces, the accuracy of the data acquisition is lower than the stand-alone 3D laser scanner; its surveying and mapping range meets the surveying and mapping needs for a small rockery less than 100 meters in length. Therefore, to meet the demands of 3D printing, this study adopts the multiple data acquisition method by utilizing a terrestrial 3D laser scanner and a hand-held 3D laser scanner to conduct data acquisition for Zhanyuan’s South Rockery. This includes the advantages of the two surveying and mapping methods, complements the data, and meets the modeling requirements. To ensure the accuracy of the scanning data while using the two scanners together, the main body of the large rockery was scanned with the stand-alone 3D laser scanner and then supplemented by the hand-held backpack 3D laser scanner. The data will be supplemented for areas whose stand-alone data cannot be collected.
The equipment used in this study was a terrestrial Trimble TX8, which has an acquisition accuracy of 1 mm, a scanning distance of 18% of the target reflectivity (above 335 m), and a measurement speed of 1,000,000 points per second; its price is $115,000 USD. The hand-held, knapsack GeoSLAM ZEB HORIZON was also used; its mapping distance can reach 100 m with a scanning speed of 300,000 points per second and an acquisition accuracy of 10–30 mm, and its price is $85,000 USD. The whole collection area of the rockery is about 2200 m2, and the large volume is covered by many plants. The main part of the rockery was collected using the Trimble TX8. Because of the intersection and shielding of the rockery’s various landscape elements, components, and irregularities during the process of erecting the rockery, to comprehensively obtain all sides of it, the hand-held scanning operation is added as well.
Because the surface of Taihu stone has an extremely complex texture and presents a porous structure, it is necessary to include multiple sites from different angles to scan every face and cranny of the rock completely. Also, because the rockery surface is covered with a large number of plants that seriously affect the actual distance of scanning and blocks the laser, an increase in the number of stands was made to ensure the complete scanning of all corners of the rockery. The total number of large-scale stations was 154 (Fig. 4). The scanning time was 3.5 h, and the total number of cloud collection points was approximately 1.66 billion. Since the narrow area and cave cannot be accessed or blocked from too much scanning information, the hand-held GeoSLAM ZEB HORIZON scanner was used to scan the details of the rockery for a short distance; however, this process does not necessitate setting up a station. The researcher carries out a full range of laser scanning to capture the object through a hand-held scanning instrument. The overall scanning time is forty minutes, and the total number of points is about 320 million.
Automatic splicing data processing without targets
Preprocessing the point cloud data followed the field data collection. Matched with the Trimble TX8, the Trimble RealWorks (TRW) software, which can preliminarily splice point cloud data, was used for processing. To avoid the problem of automatic data splicing at a later stage (due to the target being unable to set, which affects the accuracy of scanning data), the Trimble TX8 and the TRW software can realize fully automatic splicing without a target and automatically calibrate based on the degree of overlap between the cloud data from each site. Since the object of the 3D scanning is a rockery (compared to scanning the entire garden space), the research conducted for this study involves the overlapping of rocks and the surface’s texture to maintain high scanning precision.
In the process of surveying and mapping Huanxiu Shanzhuang, the spatial range was large since the research object included the entire garden; thus, the data accuracy was not high. In the mapping process of acquiring rockery data, the Leica ScanStation C10 had to maintain the instrument balance during the process of erecting the station. Since the rockery surface was mostly uneven and the site’s setting was limited, it was impossible to comprehensively scan every surface of the Taihu stone in the rockery. All of the data, therefore, could only be calibrated by the scanning target. Therefore, in the post-processed data, many holes remained in the rockery that need to be completed manually, which greatly reduces the accuracy of the research.
To ensure accurate data acquisition in all areas of the rockery during the scanning process, the number of stations at different angles was increased to 154 during the set-up process. Comprehensive data acquisition and data reconnection between the stations was achieved by ensuring point cloud calibration during the later stage. Also, the Trimble TX8 did not require an absolute level during the scanning process; thus, more stations could be set up on the rockery’s surface. Meanwhile, this instrument could cooperate with the hand-held Slam_Horizon scanner to further guarantee comprehensive data. The complete data splicing process took fifteen hours. The final average splicing accuracy was 2.67 mm, and the total number of points was 1.66 billion.
Based on aerial photos and preliminary point cloud splicing results, Zhanyuan’s South Rockery cannot be identified or studied because it is shrouded by plants. Therefore, after the point cloud preprocessing stage, it is necessary to classify the point cloud to exclude the impact that plants have on the rockery. The point clouds of the rockery and vegetation are automatically classified in the TRW software, which took an hour, approximately. Since the rockery is covered with shrubs, some are automatically identified as rockery and ground. Therefore, after the software’s automatic calculation is complete, one further step of data precision treatment is needed; the model needs to be manually refined to eliminate the impact of shrubs on the rockery while retaining the trunk and roots of some trees to avoid leaving a ground cavity. The whole model’s precision treatment time takes eight working days.
After finishing the point cloud classification, the triangle network model should be calculated next. Since the point cloud contains ample plant information, there will be model dislocation and other errors during the calculation process of the triangular network model, which need to be manually repaired. In this stage, the TRW software automatically filters the remaining shrubs to include rockery information in the model, which takes six working days. During the scanning process, the rockery was made-up of overlapping stones. Generating the software grid model would have rendered avoiding the occurrence of some small holes impossible; therefore, one step of the data precision treatment, the process is to repair the model manually, was needed. Finally, it took six working days to repair the rockery and complete the Trimble TX8 fine scanning process of the whole rockery model (Fig. 5).