ANALYZING OF THE TERRESTRIAL LASER SCANNER GEOREFERENCING USING GNSS

Recent years have witnessed emerging the cutting-edge method for point cloud creation using terrestrial laser scanner (TLS). The TLS manufacturers declare accuracies of their instruments up to the millimeter level. However, different constraints could degrade the accuracy of point cloud created by TLS. One of the obvious factors that may directly affect the accuracy of the results is a method of registration and georeferencing. In this paper, the indirect georeferencing using GNSS has been researched. The real time kinematic (RTK) technique has been suggested to measure GNSS points. The conducted test shows that average of 30 minutes data RTK-GNSS is enough to coincide with TLS data. Also, test reveals no improvements when adding more GNSS points. Nevertheless, there is an improvement in accuracy when more scans are conducted.


INTRODUCTION
The principle of TLS operation is based on the transmission of a laser beam from a TLS instrument with visible light or near Infrared which is reflected by objects and return to the instrument, and the distance (R) is determined by the time of flight (TOF) or by the phase difference. By encoders, the vertical angle (Φ) and horizontal angle (θ) are determined and combined with distance. Then Cartesian coordinates (x, y, z) of objects is obtained from distance R and angle θ and Φ as follows (Armesto et al., 2010, Reshetyuk, 2009): Where Ri, φj and θi are the measured distance, horizontal and vertical angle, respectively, to the i-th point in the point cloud, and (xi, yi, zi) are its rectangular (Cartesian) coordinates in the scanner coordinate system.
In addition, the intensity I of the reflected laser beam is often recorded which represents a fourth dimension (x, y, z, I). The result of a scan is millions of 4D points which are called point cloud.
Therefore, to benefit from the created point clouds, it should be related to known coordinate system.

GEOREFERENCING
The georeferencing is defined as the procedure of transforming internal TLS coordinate system to local or national coordinate system (Reshetyuk, 2009). Georeferencing is required if the TLS point clouds need to be integrated with other geospatial data or sequent of scans need to be related to the same system. This may be the essential step for monitoring surveying using TLS.
There are two methods for georeferencing: direct, and indirect.

DIRECT GEOREFERENCING
In this method, TLS is set up on a known point and oriented through another known point, as in Total Station (TS). Hence, the transformation parameters are set practically, i.e. the three translation parameters are determined when TLS set up and centred optically over a known point, while the rotation angles around X-axis and Y-axis are fixed through levelling procedures, finally, the rotation angle around Z-axis is set by orienting to a known point (Alba and Scaioni, 2007).
Some new generation of TLSs are integrated with other sensors, such as Global Navigation Satellite System (GNSS) and an Inertial Measurement Unit (IMU), to adopt direct georeferencing. However, this imposes additional expenses to the scanning system (Al-Durgham et al., 2014, dos Santos et al., 2013.

INDIRECT GEOREFERENCING
Indirect registration method is based on resection surveying technique to solve coordinates of station point and consequently the coordinates of all point clouds. A minimum of three known reference points is required, however, more points can be added to increase redundancy. Least Squares Adjustment is used to calculate six transformation parameters. Conventionally, with the absence of control points, surveying before scanning is required to distribute points relate to a local reference system, or to the national reference system if GNSS is used (dos Santos et al., 2013).
The indirect georeferencing is considered as the most accurate technique because the quality of results only depends on the accuracy of control points, the setting up of TLS will not affect the accuracy (Alba andScaioni, 2007, Reshetyuk, 2009). Therefore, it is selected in this paper.

THEORETICAL BACKGROUND
The principle of the indirect georeferencing, which is employed in this research, is based on three-dimensional transformation. In our proposal, the first system is GNSS coordinates (XG), while second system is scanner coordinates (XS). Hence, if there are points known in both systems, the problem is solved and any point in one system can be transformed to another easily.
This technique is known as 7-parameters transformation (Hofmann-Wellenhof et al., 2007, Reit, 1998  Consequently, if 7-parameters are known (TX, TY, TZ, µ, α1, α2 and α3), any point can be transformed between two systems. However, in our case, these parameters are unknown and to be computed from a set of points known in both systems. As far as GNSS system Since the translation vector between GNSS system and scanner is very long, and to reduce number of iterations for LSA, one GNSS point coordinates is considered as an approximate value for translation.

EXPERIMENT
To quantify the accuracy of point cloud after georeferencing, five monitoring points (numbered P5 to P9) are employed. These points are luminous stickers which can be acquired automatically 104 Hasan A. Jaafar by Cyclone software, pasted on the wall (Fig. 1). On the other hand, the georeferencing points are integration of TLS HDS target and GNSS antenna, named in this paper as TLS_GNSS target (Fig. 2). TLS model Leica HDS 3000 and GNSS model Leica GS10 were used. In addition, Real Time Kinematic (RTK) technique is suggested for GNSS measurements, position is an average of RTK measurements.
Two tests are conducted at five days apart. Each test has different constraints, as follows:

First test
 TLS was set up at two arbitrary positions, maintaining the distance to monitoring points of about 10-15 m (Fig. 3).
Kufa Journal of Engineering, Vol. 9, No. 4, October 2018 105  Four TLS_GNSS targets were used for the first and second scans. These targets were positioned ( Fig. 3) according to some criteria:  Far from building to reduce the effect of multipath.
 Distance between TLS and targets about 20m. This distance is the optimum distance for acquiring target automatically (Leica Geosystems, 2013).
 Good geometry, different directions, and different elevations.
 For each scan position, TLS acquired TLS_GNSS targets as well as monitoring points.

Second test
 TLS was set up at three arbitrary positions disregard to the positions of the first test.
 Eight TLS_GNSS points were used for the first scan, considering the criteria mentioned previously, to locate TLS_GNSS targets. Four TLS_GNSS points were used for the second and third scans.
 For each scan, TLS_GNSS targets and monitoring points were acquired.

Post-Processing
The Coordinates of TLS targets are the same as GNSS antenna, only corrected for elevation.
The RTK technique is used to measure coordinates of these targets. The average of the recorded coordinates is used (Table 1 and Table 2) For cloud points georeferencing, Cyclone7 software was used. In addition, MATLAB script is created as an alternative solution for Cyclone. The georeferencing is based on indirect technique with different constraints for each test. Table 3

RESULTS AND DISCUSSION
In order to test the accuracy of different georeferencing alternatives, the coordinates of the monitoring points are measured in two tests and the differences are computed (Table 4 and   Table 5) It can be seen that the average is the most accurate solution for Cyclone results (Fig. 7, Fig. 8, and Fig. 9), likewise results of the MATLAB script (Fig. 10, Fig. 11, and Fig. 12). For Cyclone results, in average solution, the maximum differences reach to 4 mm, 7 mm, and 10 mm for easting, northing, and elevation respectively. While for MATLAB script, the maximum differences in average solution are 4 mm, 3 mm, and 13 mm for easting, northing, and elevation respectively. Therefore, it can be inferred that the results of Cyclone software are coincide with that of MATLAB script. In addition, the accuracy of monitoring points might consider better than GNSS accuracy. This may be because by averaging of multiple scans improved the whole accuracy of the point cloud.   Fig. 8. Differences (m) in Northing coordinates for monitoring points (using Cyclone).   Fig. 10. Differences (m) in Easting coordinates for monitoring points (using MATLAB script).