|
|
Kelly C. O'Shea
|
|
In the past, the position of the eye was recorded using eye coils. The eye coil method involved surgery that was expensive and difficult. Eye coils are fragile and break easily, so surgery could have to be repeated to remove and replace them. Furthermore, the field coils interfere with other signals being recorded. The new method uses magnetic dipoles on the eye in combination with four magnetic sensors. The monkey is trained to look at a laser dot on a screen, and sensor data is recorded for several dot locations.
Four giant magnetoresistive (GMR) sensors are used to determine the strength of the magnetic field due to the magnets on the eye of the monkey. The magnetic field strength data creates a unique set of four values for every eye gaze direction. The four channels of data and the gaze direction provide a set of ordered quintets that is used as calibration data from which gaze location can be interpolated during recording. To go from four channels of raw data and locations of laser spots on a wall, the methods of Delaunay tessellation and meshing are used.
Delaunay tessellation is a way of connecting all given points in space with simplicies so that no other point lies in the circumsphere of any simplex. A simplex is a geometric volume, in this case in 4-space with five vertices and five faces. This construction is useful because it creates a set of points where each point is connected only to its nearest neighbor and can be used uniquely for creating a surface with minimal discontinuity. The tessellated space is useful in the next step, meshing. Meshing is taking a set of training data and fitting a curve or surface to it. In this case, the program MATLAB uses Delaunay nature of the tessellation of the data to find a 4-D surface with the least error possible.
The training sensor data is graphed in four dimensions, and the Delaunay method is used to tessellate the 4-D space. Each data point is used as a node, and the known gaze directions from the training of the monkey are used as the fifth variables during meshing.
Eye motion occurs as a combination of three types of movement: horizontal, vertical, and torsional. To determine gaze direction however, only the horizontal and vertical movement is needed. The known gaze direction is recorded as an ordered pair of two angles. The sensor data is meshed two times, once with each variable of direction. After meshing, the two five-dimensional surfaces are used to interpolate data from recording with the result being a pair of angles the eye was rotated horizontally and vertically. Once the training data has been meshed, the resulting surfaces are used to interpolate the sensor data taken from recording during experimental paradigms.
|
|
|