Quality Assessment of 3D Point Clouds on the Different Surface Materials Generated from iPhone LiDAR Sensor

Quality Assessment of 3D Point Clouds on the Different Surface Materials Generated from iPhone LiDAR Sensor

DOI: https://doi.org/10.52939/ijg.v18i4.2259

Laser scanning is a common land surveying tool for accurately measuring and collecting data from artifacts, structures, buildings, and landscapes. This emerging instrument is incrementally used in the AEC industries which improve the productivity and management efficiency of several field disciplines (Telling et al., 2017,). The principle of laser scanning was defined by Böhler and Marbs (2002) in which the laser lights are used to measure distance from the sensor to the object. It operates by pulsing a ray of light that rotates vertically 270 degrees and horizontally 360 degrees. The light shall be mirrored backwards when any surface touching and registered as a data point that assigns a reflection of colour and surface. To be able to read the data, the 3D point clouds which are composed from millions of 3D coordinates (XYZ coordinates) could be used to provide a precise and detailed image of the environment.

In this study, two datasets were applied from Terrestrial Laser Scanning (TLS) and iPhone LiDAR sensor. Generally, TLS is a well-established method for collecting 3D data on a small scale (Jaboyedoff et al., 2012, Riquelme et al., 2021, Suchocki and Katzer, 2018 and Wu et al., 2021). TLS, also known as terrestrial LiDAR, determines the XYZ coordinates of numerous locations on the ground by delivering laser pulses to all spots while measuring the distance between instrument and the target. Classifying terrestrial laser scanner is possible on the basis of the measuring concept (i.e. triangulation, step, pulse) or the technical criteria obtained (Frohlich and Mettenleiter, 2004). For all possible uses, there is no universal laser scanner.

Certain scanners are well-suited for use in indoor as well as medium-sized operations. Other scanners with a longer range (up to 100 m) are more suited for outdoor use, although close-range scanners with excellent precision are also available (up to a few metres). Multiple geodetic measurements are the most popular use of the models generated with TLS (Spreafico et al., 2021).

Recently, the generation of 3D point clouds using smartphones have been introduced since the last decade. According to Sirmacek and Lindenbergh (2014), the production of point clouds using iPhone 3GS smartphone sensor data might be a quick, low-cost, and less complex as compared to TLS, which then could enhance 3D reconstructed model by incorporating a smartphone-generated point and conduct several assessments on the quality of 3D point clouds in terms of density, point-to-point distance and local roughness values between iPhone and TLS. In our study, the focuses are on the evaluation of the density of 3D point clouds and the accuracy of distance measurement from mesh 3D point clouds at different surface materials generated from the iPhone LiDAR sensor and comparison to TLS was for a reference data.

2. Methodology

The general methodology for this research starting from the planning and area selection, data collection and processing, result and analysis until to conclusion and recommendations. Figure 1 shows the general methodology for this study.

2.1 Project Planning

In this stage, the selection of study area, materials, instruments, processing software and research design were decided to ensure the study run smoothly. Such conditions on study area, types of instruments and surface materials plus the software selection have been considered for an ideal condition at the standard operating procedure when conducting the laser scanning task.

Figure 1 General methodology

Figure 2 The view in B516 Classroom

This device is able to capture 3D point clouds up to 2 million points per second at the range accuracy of 1.0 mm + 10 ppm. Also, this scanning device has different 3D point accuracy which are divided into three classes ranging from 10m, 20m and 40m at 1.9mm, 2.9mm and 5.3mm respectively.

On the other hand, the iPhone 12 Pro uses a scanner-type LiDAR sensor which applied flash illumination and no scanning method for data acquisition. With the integration of 8XX nm wavelength, Photon Counting detectors or known as SPADs or Single Photon Avalanche Photodiodes, plus the Vertical Cavity Surface Emitting Lasers (VCSELs), this device is able to capture point clouds in a short period of time, but has a limited Field of View (FoV) which at ~ 5m range distance between sensor and object and perfectly adequate for the types of consumer applications. Due to these specifications, it is a good thing to have several tests on the quality obtained from the 3D reconstruction at different surface materials within 5m range of scanning scenario where the main focuses are given to the density and the accuracy of distance measurement gathered from the 3D point clouds.

Table 1 Types of object materials

Lastly, all the datasets output from both scanning sensors were uploaded into CloudCompare software which was used to view the point cloud datasets from both sensor in one screen as well as creating 3D reconstruction model, cropping process, obtaining point cloud density and measuring the distance lines for each surface material.

3. Result and Analysis

3.1 Point Cloud Density for Aluminium

Referring to Figure 3, the 3D reconstruction object for aluminium on both instruments produced an actual shape after the mesh process applied. The mesh of 3D reconstructed model from TLS produced a real good surface while some black spots were presented in iPhone LiDAR sensor which indicates some missing point clouds that did not reflected back to the sensor. The density of point clouds using TLS and iPhone LiDAR sensor was measured at 469,856 points and 9745 points respectively in 1m² of dimensional area.

3.2 Point Cloud Density for Plastic Board

Plastic board is an object that has a low reflectivity, high in absorption, dull in colour and less tough than other surface materials. These made the distribution of point clouds unevenly scattered. It is dull in colour and less tough than other surface materials. As the comparison, the meshing process going smoothly which resulted in almost similar model in both iPhone LiDAR sensor and TLS instrument (Figure 4). All the white lines of reflection can be seen in both images and it is a quite good achievement for the iPhone LiDAR although the black spots can be seen clearly. The point clouds density for TLS is 380,612 while 9928 for iPhone in 1m² of dimensional area.

3.3 Point Cloud Density for Canvas

Referring to Figure 5, the 3D meshes for canvas in iPhone LiDAR was successfully reconstructed at the actual object shape appearance. However, the blurred image of center point mark in the 3D meshes was identified for both datasets, especially for the iPhone LiDAR sensor. This is due to a high absorption surface that resulting to the slower rate of reflected points reached the instrument. Meanwhile, the density of 3D point clouds in the TLS and iPhone were calculated at 112,410 points and 845 points respectively in 0.09m² of dimensional area.

Figure 3 Point clouds distribution for Aluminium using TLS (left) and iPhone (right) in CloudCompare Software

Figure 4 Point clouds distribution for plastic board using TLS (left) and iPhone (right) in CloudCompare Software

Figure 5 Point clouds distribution for Canvas using TLS (left) and iPhone (right) in CloudCompare Software

3.4 Point Cloud Density for Plywood of Board

Figure 6 showed the object reconstruction for plywood of board marker. It can be seen that the 3D meshes of iPhone LiDAR sensor have a close real shape even there was some black spots representing the missing point clouds. However, most of the point clouds are constantly reflected back to the instrument. For point clouds density, the TLS and iPhone LiDAR sensor were recorded 194,167 points and 3,539 points respectively in 0.36m² of dimensional area.

3.5 Point Cloud Density for Plywood

Referring to Figure 7, the 3D meshes of plywood was closely reconstructed at real shape. It was discovered that the laser beam transmitted precisely on the object surface and reflected well to the sensor. Due to the absence of moisture on this plywood surface, the point clouds were reflected in a consistent direction. The density of 3D point clouds for TLS and iPhone LiDAR sensor was recorded 416,359 points and 10,958 points respectively in 1m² of dimensional area.

3.6 Point Cloud Density for Ceramic Tile

For this type of surface, the 3D meshes presented from the iPhone LiDAR sensor didn’t well reconstructed due to less reflected point clouds similarly to canvas surface (Figure 8). The color of black (low reflectance) and white (high reflectance) in one object surface was also contribute to less quality of 3D point clouds output which deal with 3D point clouds density. For the point cloud density using TLS, it is recorded as much as 42,720 points while 859 points for iPhone in 0.09m² of dimensional area. As overall, the 3D point density for all types of surface materials can be concluded as shown in the Figure 9. Referring to the TLS datasets, the densest point clouds was the aluminium, plywood and plastic board as regard to the 1m² dimensional area, followed by plywood board marker (0.36m ²), canvas and ceramic tile for 0.09m². Even tough the difference in dimensional area, the values are still reasonable if equalized with percentage. This result is the benchmark for 3D point clouds dataset as the TLS provides the high accuracy of laser scanning measurement. However, different result has been obtained for the iPhone LiDAR sensor where the densest point clouds was the plywood, plastic board, and aluminium for 1m² dimensional area, followed by plywood board marker (0.36m²), while both ceramic tile and canvas which at 0.09m² dimensional area have a slightly difference of the number of 3D point cloud density.

Figure 6 Point clouds distribution for plywood of board marker using TLS (left) and iPhone (right) in cloud compare software

Figure 7 Point clouds distribution for plywood using TLS (left) and iPhone (right) in

cloud compare software

Figure 8 Point clouds distribution for ceramic tile using TLS (left) and iPhone (right) in cloud compare software

Figure 9 Density of point clouds for each surface material between TLS and iPhone LiDAR sensor

Table 2 Root Mean Square Error (RMSE) for different surface materials

4. Conclusion

As the conclusion, this study managed to compute meshing process in both Leica RTC360 laser scanning and iPhone 12 Pro LiDAR scanner using the suitable software, CloudCompare. By using the result in CloudCompare software, the density histogram of 3D point clouds has been used to get the information of the density of 3D point clouds depending the dimensional area per surface materials. It can be seen that the factors of roughness, reflectivity strength, colors, texture and type of surfaces were influenced the reaction of the 3D point clouds when reaching the surfaces (Hezri Razali et al., 2021). This has been proved from the number of 3D point clouds at each of surface materials as discussed earlier. Next, this study also managed to identify the capability of iPhone 12 Pro LiDAR Sensor in measuring feature dimensions from different object materials as compared to ideal measurement using Terrestrial Laser Scanner (TLS). The RMSE for the data in both instruments have been calculated using the right formula in order to determine the best object surfaces that has the least error which close to 0 value and the denser point clouds has resulted a good accuracy of distance measurements using 3D mesh images. It showed that the iPhone LiDAR sensor is capable to provide a better accuracy of below than 50cm within a small space of indoor environment, approximately at 100m ² room area. This research also succeeded in identifying the interaction between point clouds and the properties of the surface materials which influence the diversion of different point clouds emitted and reflected back to the scanning sensor.

References

Böhler, W. and Marbs, A., 2002, 3D Scanning Instruments. Proceedings of the CIPA WG 6 International Workshop on Scanning for Cultural Heritage Recording , 9-12, http://www.i3mainz.fh-mainz.de/publicat/korfu-/p05_Boehler.pdf.

Frohlich, C. and Mettenleiter, M., 2004, Terrestrial Laser Scanning - New Perspectives in 3D Surveying. International Archives of Photogrammetry.Remote Sensing and Spatial Information Sciences, Vol. 36, 7-13.

Jaboyedoff, M., Abellán, A., Derron, M. H., Loye, A., Metzger, R. and Pedrazzini, A., 2012, Use of LIDAR in Landslide Investigations: A Review. Natural Hazards, Vol. 61, 5–28. https://doi.org/10.1007/s11069-010-9634-2.

Hezri Razali, M., Idris, A. N. and Nor, T. N. A. T. M. and Ghazali, R., 2021, Accuracy Assessment on Point Cloud Dataset from Terrestrial Laser Scanner with Different Objects Surface Properties . IOP Conf. Ser.: Earth Environ. Sci., Vol. 767, https://doi.org/10.1088/1755-1315/767/1/012007.

Martínez-Olvera, C. and Mora-Vargas, J., 2019, A Comprehensive Framework for the Analysis of Industry 4.0 Value Domains. Sustainability, Vol. 11(10), 1–21. https://doi.org/10.3390/su11102960.

Riquelme, A., Tomás, R., Cano, M., Pastor, J. L. and Jordá-Bordehore, L., 2021, Extraction of Discontinuity Sets of Rocky Slopes Using iPhone-12 Derived 3DPC and Comparison to TLS and SfM Datasets . IOP Conference Series: Earth and Environmental Science, Vol. 833(1). https://doi.org/10.1088/1755-1315/833/1/012056.

Sirmacek, B. and Lindenbergh, R., 2014, Accuracy Assessment of Building Point Clouds Automatically Generated from Iphone Images. International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences - ISPRS Archives , Vol. 40(5), 547–552.https://doi.org/10.5194/isprsarchives-XL-5-547-2014.

Spreafico, A., Chiabrando, F., Teppati Losè, L. and Giulio Tonolo, F., 2021, The Ipad Pro Built-In Lidar Sensor: 3D Rapid Mapping Tests and Quality Assessment. International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences - ISPRS Archives , Vol. 43(B1-2021), 63–69. .https://doi.org/10.5194/isprs-archives-XLIII-B1-2021-63-2021,

Suchocki, C. and Katzer, J., 2018, Terrestrial Laser Scanning Harnessed for Moisture Detection in Building Materials-Problems and Limitations. Automation in Construction, 94(June), 127–134. https://doi.org/10.1016/j.autcon.2018.06.010.

Telling, J., Lyda, A., Hartzell, P. and Glennie, C., 2017, Review of Earth Science Research Using Terrestrial Laser Scanning. Earth Sci. Rev., Vol. 169, 35–68. ,https://doi.org/10.1016/j.earscirev.2017.04.007,.

Wu, C., Yuan, Y., Tang, Y. and Tian, B., 2021, Application of Terrestrial Laser Scanning (TLS) in the Architecture, Engineering and Construction (AEC) Industry. Sensors, Vol. 22, https://doi.org/10.3390/s22010265.