saccade | sophisticated scientific image viewer for Linux with OpenGL | Computer Vision library

If there are more than one column, use filter_at

The key issue here is the usage of polar coordinates.

The densities must be computed on xy cartesian coordinates, usage of polar coordinates results in weird distance values, try plotting your data in a rectangular plot and will see.

You can include a coordinate transformation before the z values are computed. E.g.:

Before you attempt to analyse your data, you must visualise it. This will help you see if the data is actually what you think it is, and to show qualitatively the extent of some data quality issues.

If your data is actually a gaze data time series, then you could visualise it as in your example above simply by plotting x as a function of time and y as a function of time. This would not require that you run the data through a saccade detection algorithm first. The plot you show above is simply a visualisation of the raw data, with the saccade detections superimposed (the grey lines), but that is an optional step.

Now to your raw data: if you plot it as suggested, I suspect it isn't going to be what you think it is. Compare it to the example data that you posted from https://github.com/tmalsburg/saccades. That dataset shows a couple of characteristics:

• The raw data is given with sub-pixel precision (i.e. to 2 decimal places). This is typical of calibrated eye data, where the estimated gaze signal at any point is the output of a calibration process that maps coordinates from the original eye video data to the viewed image coordinates, via a continuous function that yields interpolated values that appear to exceed the resolution of those images. Your data, however, consists of integers, which is unusual (but possible).
• More importantly, the Github example data also shows that each sample position varies very little from its preceding one (e.g. the x position shifting from 53.18 to 53.20 from one sample to the next). This is typical of real gaze data during a fixation. During a saccade, adjacent samples start changing much more rapidly of course, and you can often detect them visually simply by scrolling down a column of values and noting where the adjacent values change quickly. Compare this to your data, which is completely different: values bounce up and down erratically from one sample to the next (e.g. from 732 to 706 to 666 to 886 in successive samples). This is not typical of a good gaze data signal. So plot your time series and see what you actually have there.

I suspect that there are discontinuities in your data, and this will cause any saccade or fixation detection algorithm to fail. Try to figure out what your data actually is representing, and/or what the quality issues are, before attempting to parse into saccades and fixations.

Hint: using the ggplot2 library to visualise your data:

Try ungroup :

I'm not sure if I understood your request correctly. The time between the first occurence of block == 'target.1' and the first occurence of block == 'target.1' & saccade == 'S' for each trial could be calculated like this:

We can create an index using rle() and then group_by() this index to sum() the time:

You could do this using dplyr.