-
Notifications
You must be signed in to change notification settings - Fork 4
Reduce Noise
noise_reduction R Documentation
This function reduces noise in the eye tracking data by applying the specified method (usually either median or mean) over the specified window
noise_reduction(data, method = median, window_size = 3)
data Data frame of the eye tracking data we want to process
method function which is given the current data window and returns the denoised value (usually either median or mean)
window_size Number of data points over which we want to apply the noise reduction
Input data frame columns
gazeorigin_x
: X coordinates of the gaze origin
gazeorigin_y
: Y coordinates of the gaze origin
gazeorigin_z
: Z coordinates of the gaze origin
gazeDirection_x
: X coordinates of the gaze direction
gazeDirection_y
: Y coordinates of the gaze direction
gazeDirection_z
: Z coordinates of the gaze direction
gazePoint_x
: X coordinates of the gaze point
gazePoint_y
: Y coordinates of the gaze point
gazePoint_z
: Z coordinates of the gaze point
This is an implementation of the noise reduction function as documented in the Tobii Pro I-VT fixation filter. It goes through all gaze points and replaces each data point by the result of the specified method (in case of Tobii either median or mean) over all data points in a window centered around the data point currently being replaced.
As the eye tracking rate of the HoloLens 2 is 30 Hz, a noise reduction over 3 samples would result in a window size of 1/30Hz = 100ms. With the typical fixation duration around 200ms this would result in a window not much smaller than the fixation duration and risk significantly influencing the data negatively. Therefore it should not be used by default.
The modified input data frame with the noise reduction applied and an additional column modified which indicates if the line was modified