The most direct approach to extract 3D digital data of cultural relics is laser scanning measurement, which is featured by fast speed and high precision without secondary damage to cultural relics, so as to record and preserve the original information of cultural relics to the greatest extent[8]. In terms of the Qing Dynasty double-ear clashing color vase with lotus design, this research obtained the 3D digital model of porcelain bottle with the help of laser 3D modeling technology, and mapped the texture information. Finally, 3D printing was completed to realize the virtual reproduction of double-ear clashing color vase with lotus design. The modeling process was based on the RapidForm XOS2 laser data processing software, the collected point cloud data were automatically spliced to obtain the 3D model of cultural relics with texture information, and then combined with reverse engineering software Geomagic Studio was, the missing part was patched or scanned, and the texture image was used to make up for the defective texture, which eventually ensured the acquisition of 3D data model of cultural relics. Currently, 3D scanners are divided into two categories, one is airborne (LiDAR) type, which is mainly aimed at large ground scenes such as cemeteries, and the scanning accuracy is relatively low. The other is terrestrial type (TLS), which is mainly used for small objects[7.9]. Compared to LIDAR, TLS scanning is of higher accuracy and easier operation. This research has chosen the TLS for 3D scanners. The principle of TLS is to veritably reproduce the three-dimensional landscape of the measured object by transmitting and receiving pulsed laser to collect point cloud data. However, it is noteworthy that as to porcelain, the ground laser scanner is incapable of obtaining complete and uniform point cloud data due to disparate angle reflection caused by light. Therefore, in addition to FreeScan X3, contacting three-dimensional laser scanner, NikonD850 digital camera for cultural relic data collection and high-resolution texture information extraction, the process of obtaining three-dimensional data also employed software such as 3DMAX,CINEMA 4D fir denoise and elimination of redundancy, then extracted the three-dimensional point cloud data of double-ear clashing color vase with lotus design[10.14.16]. The 3D data acquisition process consists of three procedures: data acquisition, data editing and data storage, as
the detailed flows shown in Figure 1.
Fig. 1 Flow chart of point cloud data processing
2.1 3D data collection
The data acquisition includes three aspects, namely, regular equipment debugging, data acquisition, point cloud data preprocessing and storage. For data acquisition, a quick look 3D model was obtained by rough data processing to ensure the integrity of data. For data processing, the laser data optical data obtained from 3D model were verified each other to make sure the 3D accuracy of the results. The inserting value calculation was conducted to the data at all levels, to count the accuracy and diminish the accuracy loss in data processing. Lastly, the fusion processing of texture information obtained by optical camera and texture information in laser data guaranteed the texture accuracy of the final model[11.18.19.20].
Equipment debugging refers to the sensor equipment, such as laser scanner and digital camera, used in the experiment, which is affected by the transportation, storage environment or other conditions, causing deviation in accuracy. To ensure the accuracy of cultural relics data acquisition, the sensor needs to be calibrated before application[12.13.15].
After completing the sensor debugging is to carry out data acquisition. The high-density laser data collected from double-ear vase and data-entry mainly operated by FreeScan X3 scanners and NikonD850 digital cameras. The cultural relic was placed on the rotating platform; the appropriate measuring station position was selected according to the size of the double-ear vase, and the scanning density parameter was set to the sampling point of each scanning laser line in760 points / line. The relative position of the sensor and the cultural relic was adjusted in line with the lens parameters of the sensors, then the two sensors collected data in the directions of the transverse R(rotation of platform) and longitudinal H (height), respectively. In the meantime, adopting 3D laser scanner and digital camera to collect 3D coordinates and texture information of cultural relics, the number of points obtained is 286578, and the average point spacing is about 0.2mm.
In 3D data acquisition, the selection of coincidence rate is crucial to the quality and efficiency of the model. Several groups of point cloud data can be obtained after multiple scans of a complete 3D model, the higher the coincidence rate of these groups of data, the more accurate the data acquisition, the better the effect of the model after the later splicing, and the lower the efficiency of data acquisition. On the contrary, if the coincidence rate is too low, even if the extraction is highly efficient, later it may result in large amounts of errors in the calculation of point cloud data, causing large broken surface in the model[19]. Therefore, data acquisition needs to strike a balance between the quality and efficiency, to intentionally select proper coincidence rate. During the data acquisition process, the vase was placed vertically on the turntable, and a 30° was rotated each time to obtain one-stop data. The 360° rotating was defined as R, then the data lift scanner acquired in 360° rotation was defined as H. R scanning only required turning the turntable, given it did not involve the position coordinates movement of the scanning device, the extraction image is of high coincidence rate, while H needed to adjust the height of the scanning device up and down, and the absolute position of its X、Z axis would also change after moving the scanner Y axis, so the extraction image repetition rate was relatively low. Since the later splicing process began with transverse direction before longitudinal direction, though the longitudinal coincidence rate was small, the data involved in the splicing were fixed as the data between the two strip transverse acquisition bands, so it can ensure the accuracy of the splicing. The final data comparison shows that, as long as R≥75%, HR≥45% the
data coincidence rate, it can meet the requirements. as shown in Figure 2.
Fig. 2 The schematic diagram of 3D data’s collection on cultural relic
The last procedure was point cloud data preprocessing and storage. Data preprocessing is to check the quality of data acquisition. In order to improve the efficiency of model acquisition, plus the necessary gaps in scanning process, the initial point cloud data will inevitably be incomplete. The missing point cloud data, reflected in the 3D model, is the broken surface of the model. To this end, it is necessary to preprocess the collected data in the field of data acquisition for the integrity of the collected data.
Above all, the obtained pre-processed 3D point cloud data should be further refined. The initial point cloud data acquired by 3D scanning is liable to be rough, which is suitable for the 3D splicing of some complicated ornaments, such as porcelain. To solve this problem, three stages are needed, including laser data processing, digital photogrammetry processing and data fusion. Laser data processing refers to the automatic and manual trimming of point cloud data obtained by each scanning station, and the registration processing of all point cloud data of porcelain, in which obtain the connection points of all two coincidence data and coincidence areas, and then connect the points to form a spatial transformation equation. Subsequently the point cloud data is converted into a plane by computer automatic settlement.
Considering the collected data will appear partial overlap, the data calibration must be carried out by manual calibration, and ultimately make gridding for the whole model after calibration to obtain the quadrilateral model. Owing to the four sides of the whole model, the four sides grid reconstruction was carried out in accordance with the curvature geometric characteristics of the double-ear vase surface and the automatic calculation method of the software, to obtain the micron-scale 3D model in four sides, and finally, the high-precision three-dimensional model was realized followed by model grid optimization.
2.2 Digital photogrammetry
Digital photogrammetry processing is to determine the relationship between images by human-computer interactive editing, and then obtain dense connection points by R, H image registration for dense connection points. In addition to the accuracy and integrity of basic data, the degree of color restoration, the difference of image data and acquisition data in texture accuracy and photo color also have direct impact on the data quality of 3D reconstruction. NikonD850 was selected in this digital photogrammetry, it is also the current mainstream photogrammetry camera, with CCD sensor size of 35.9*23.9 mm, the highest resolution 8256*5504, and the effective pixels 45.75 million, after installing the MD-18 handle, the NikonD850 can reach about 9 beats per second in continuous spots, which can fully support the texture information collection. NikonD850 camera can start to collect the texture information of the double-ear vase, after correcting the color card, then calculate the camera focal length and determine the shooting position in the resolution of no less than 300 dpi. At the premise of keeping the focal length and shooting distance unchanged, the coincidence rate of the upper and lower, left and right adjacent was not less than 45% and 75%, respectively, then the target object was photographed. Finally, the 76 sequence photos of no less than 300 dpi were extracted to form the texture information of the cultural relic. Four pairs of points of the same name were selected as the control points on the sequence photos and the obtained high precision 3D model. the (x,y) coordinates of the same name points on the photos and the (X,Y,Z) coordinates on the 3D model were recorded for registration to acquire the 3D model with texture information.
Considering the influence of ambient light on the double-ear vase, all the images were processed in uniform color. Due to the different light intensity on each side of the vase, the uniform light source should be set before adjusting the texture information. As for color texture information, stable light source serves as a vital component. After laser data processing, a 3D model with high precision artifacts was obtained. However, the texture information cannot be directly mapped to the 3D model, because the chromaticity of the same area will produce different effects due to different angles of light, that is, the same vertex of the cultural relic grid model is given different color values. In this case, direct mapping of texture information will result in unidentical texture information calculation. Hence, light source in uniform brightness and saturation should be set around the cultural relic before the texture information extraction, to create a uniform color temperature and unified light source, and then adjust the texture picture[16]. Table 1 is a reference of light source and projection relation of 3D reconstructed artifacts, among which, "Spatial mean resolution (mm)" is a key index to identify the accuracy of texture mapping data, and "Projection error" is
a crucial index to judge the accuracy of projection matching.
Table. 1 Reference of light source and projection relation of 3D reconstructed artifacts
By digital photogrammetry, the data with high accuracy and rich texture information will be obtained. But the data points are relatively dense, which requires data fusion processing. The purpose is to inspect the overall 3D precision of the model and supplement the texture information of the cultural relic model, so that enable the shape and color of the double-ear vase to reach the effect closest to the entity.