sarmadm / saccades

Detection of fixations in eyetracking data

Saccade and Fixation Detection in R

News

• 02/16/2015: saccades is now available on CRAN.

Overview

An R package for saccade and fixation detection in eyetracking data. It uses the algorithm for saccade detection proposed by Ralf Engbert and Reinhold Kliegl. Any period occurring between two saccades is considered to be a fixation.

This software is a re-implementation of an earlier package that we used heavily in our research. However, I wrote the old package when I was new to R and the code was somewhat clumsy. Hence, the rewrite.

Note that the new code is not tested as much as the earlier package and it may have bugs. For my data it works beautifully but your mileage may vary.

Things that I plan to add in the future are tools for reading common file formats for eyetracking data and tools for assessing the quality of the fixation detection. A lot of this code already exists in the old package and I will migrate it in small steps. For example, we have a highly efficient parser for EDF files that uses the edf-library provided by SR Research.

Build and install package

These instructions assume that you are on a Linux or OSX system. To build the package execute the following command in the root directory of the git repository:

make build

This creates the file saccades_0.1.tar.gz which can be installed with this command:

sudo R CMD INSTALL saccades_0.1.tar.gz

Getting started

If you want to play with the package, this is how you get started:

> library(saccades)
> data(samples)
> head(samples)
	  time     x      y trial
1    0 53.18 375.73     1
2    4 53.20 375.79     1
3    8 53.35 376.14     1
4   12 53.92 376.39     1
5   16 54.14 376.52     1
6   20 54.46 376.74     1
> fixations <- detect.fixations(samples)
> head(fixations[c(1,4,5,10)])
  trial        x         y  dur
0     1 53.81296 377.40741   71
1     1 39.68156 379.58711  184
2     1 59.99267 379.92467   79
3     1 18.97898  56.94046  147
4     1 40.28365  39.03599  980
5     1 47.36547  35.39441 1310

If you want to examine the results of the saccade detection visually, you can use the function diagnostic.plot:

> diagnostic.plot(samples, fixations)

This function will open an interactive plot showing the original samples and the detected fixations. The plot can be used to navigate the whole data set using the mouse or keyboard. Here's a screenshot:

The dots are the raw samples. Red dots represent the x-coordinate and orange the y-coordinate. The vertical lines mark the on- and offsets of fixations. The horizontal lines (difficult to see in the screenshot) represent the fixations.

The function calculate.summary prints some summary statistics about the detected fixations:

> stats <- calculate.summary(fixations)
> round(stats, digits=2)
                               mean       sd
Number of trials              10.00       NA
Duration of trials         37015.70 16513.25
No. of fixations per trial    97.00    44.22
Duration of fixations        340.00   458.29
Dispersion horizontal          2.60    13.16
Dispersion vertical            2.31     8.41
Peak velocity horizontal       0.06    22.18
Peak velocity vertical        -0.16    13.53

Blinks

The package currently doesn't offer blink detection. However, blinks are fairly easy to spot. In the graph below you can see a blink. It starts with something that looks like a saccade, then there's a fixation on the coordinates (0,0), and then another saccade ((0,0) is what SMI eyetrackers give you when there was track loss).