Biomedical Engineering Reference
8.2.2 Evaluation of the FaceAPI for Direct Head Tracking
For evaluation of the FaceAPI for application in neuro-navigated or robotized
TMS, we first mount a human head phantom to the robot's end effector and
position a webcam opposite to the robot [ 8 ]. As webcam we use a Logitech Quick
Cam Sphere (Logitech international S.A., Morges, Switzerland) with a maximum
resolution of 1600 1200 pixels. We now move the robot with a given pattern and
track the head phantom with the FaceAPI using the webcam images. To validate
the tracking data with respect to the current system, we place a Polaris tracking
system next to the webcam and attach a headmarker to the phantom. Subsequently,
we track the head phantom with the Polaris system. For evaluation, we compute
the deviation of the FaceApi tracking results to the tracked data by the Polaris
8.2.3 Accuracy of the FaceAPI
We have found that the translational error is 9.99 mm with an Standard Deviation
(SD) of 3.76 mm when using the Polaris as ground truth. The maximum error is
21.6 mm. The rotational errors are 1 : 68 ,1 : 60 and 1 : 99 for the x-, y- and z-axis,
respectively, with standard deviations of 1 : 12 ,1 : 35
and 1 : 39 . The maximum
rotational error is 10 : 3 [ 8 ].
As this error is too large for the application of TMS, we seek for another
technique for direct head tracking.
8.3 3D Laser Scans
Previous investigations have shown that a three-dimensional laser scanner is
suitable for 3D recordings of the human face [ 12 ]. Thus, 3D laser scanning sys-
tems are one option for direct head tracking. These systems are well established in
medical applications. The main application is in gating for radiotherapy and
Computed Tomography (CT) [ 10 ].
A laser scanner measures the surface of an object, in this case the skin surface,
in relation to the scanner. For this purpose, a laser beam is moved column or grid
based over the surface to scan the contour. The scanner then measures the
deflection of the sent beam to compute the surface. In this way, a 3D surface is
provided by the laser scanner.
Direct head navigation for robotized TMS using laser scans is based on three
steps: First, a calibration from laser scanner to robot must be performed. Second,
the TMS coil (or tool in general) is registered for precise positioning and stimu-
lation. And in the third step, the laser scanner is used for position acquisition.