Direct Head Tracking - Robotized Transcranial Magnetic Stimulation - page 136

8.3.2 Data Acquisition for an Experimental Validation
After presentation of the steps required for direct head tracking, we perform an
experimental evaluation of the presented method. Therefore, we test each of the
required steps separately with the 3D laser scanner. As an accurate calibration is a
basic requirement for the application of 3D laser scans for direct head tracking in
the robotized system, we foremost estimate the accuracy of the robot to laser
scanner calibration. Subsequently, we evaluate the accuracy of the head tracking
based on 3D laser scans. An accurate registration of the 3D laser scans to the
reference image is a prerequisite for accurate tracking results. Therefore, we obtain
laser scans of a human head phantom and evaluate the accuracy of the registration
to a reference image that was generated from an MRI-scan. For estimation of the
accuracy of head tracking based on 3D laser scans, we mount the human head
phantom to the robot. We move the head phantom with the robot to different
positions and record a 3D laser scan at each position. We now use the robot
positions as ground truth to estimate the tracking accuracy. Finally, we present 3D
laser scans of two typical TMS coils that can be applied for coil calibration.
8.3.2.1 Calibration
As described above, using the laser scanner is not as simple as using an off-the-
shelf tracking system. It is time consuming to collect a large set of data points for
calibration. For calibration, as described in Sect. 8.3.1.1 , we mount the calibration
tool to the robot's end effector and utilize this tool as a marker for the laser
scanning system.
Consequently, we use a set of only n ¼ 50 randomly distributed data points to
test calibration of the laser scanning system. Beside the QR24 algorithm, we also
test the QR15 algorithm and an extended version of the QR24 algorithm, called
QR24 M , which uses a scaling factor of 0 : 001 for the translational part of the
calibration matrix. Additionally, we test the two classical methods for hand-eye
calibration by Tsai and Lenz [ 17 ] and the dual quaternion approach [ 6 ]. See also
Sect. 4.1 for an overview of the methods for solving the hand-eye calibration
problem.
We test the calibration methods with 5 ; ... ; 25 data points. We use the
remaining 25 data points to verify the estimated calibration matrices. For evalu-
ation, we apply the calibration matrix and compare the transferred pose of the
marker to the recorded pose (by the robot). As error measure, we use the absolute
translational error and the absolute rotational error.