There are many methods to identify saccades and determine onset and offset. This is what I used in Smeets, J. B. J., & Hooge, I. T. C. (2003). Nature of variability in saccades. Journal of Neurophysiology, 90, 12-20.

Saccade detection was done by a Matlab program that marked saccades by a velocity threshold of 75°/s. After detection of all saccades, the program searched for onsets and offsets of the saccade. First it determined the averaged absolute velocity of a 100-ms period of fixation that started approximately 200 ms preceding the saccade. Onsets and offsets were determined by searching the time when absolute saccade velocity reached a value 3 SDs higher than the absolute velocity during the fixation.

Hey Patrick,

in general saccades are distinguished from other eye movements by their stereotypical relationship between amplitude and duration/peak velocity, e.g. the main sequence (MS; see Bahill/Baloh). Since you are at the low end of the sampling rate and peak velocities are VERY sensitive to that, I would try to use the amplitude-duration MS to determine which saccades are artifactual. If you have enough saccades you can fit a linear regression (the A-D MS is linear over a large range) and identify saccades outside your prediction interval.

Be aware that this method has some difficulties:

1. It does only take care of strong outliers (e.g. your 2000ms saccade).

2. it is less sensitive for small amplitudes, as the differences in durations are smaller.

3. It is less sensitive at higher noise levels and to the ability of your event detection method to determine the saccade duration correctly (and dispersion-based methods will introduce more noise in the measured durations).

4. It should only be used for healthy, middle-aged populations as the MS might be altered in very young/old or pathological subjects.

5. The same holds true for different saccade types (e.g. cognitively- vs. reflexively-driven) as these follow different MS relationships.

Hope this helps!

Greetings, David

Article Quantitative Measurement of Saccade Amplitude, Duration, and Velocity

Article The main sequence, a tool for studying human eye movements

It's going to be very difficult to identify small saccades with 50 Hz sampling as you simply won't have enough samples occurring during the saccade. Check out Nystrom & Holmqvist, An adaptive algorithm for fixation, saccade, and glissade detection in eyetracking data Behavior Research Methods, 2010, 42 (1), 188-204 for a modern discussion of this problem. If your saccades are >5 deg you should be able to see them but below that they may get lost in the noise.

Hello. There are already good answers to your question from Jeroen, David and Larry above. To these, I would only add that detection of any time-varying signal is only as good as the statistical model on which it is based. From a brief reading of the above, I would suggest looking at more than just the estimated variance (dispersion) of the signal with corresponding estimation of onset and termination times. In my experience, Markov models are usually very effective for many physical phenomena since they provide a reasonable model of the time dependence. I would therefore suggest looking at a probabilistic detector structure that incorporates a Markov or perhaps semi-Markov signal model for saccades. The other thing to look for is whether you need to perform detection in real time or whether you can wait till more data are available. Post-processing (using smoothing techniques) is a good way to establish truth data since it is inherently more accurate than online filtering.

Hello, by any chance, do you known a simple matlab toolbox (or just a function) that can clean raw eyetrackerdata, and detect saccade onset et offset based on one of these method? Thanks