Automated classification and scoring of smooth pursuit eye movements in the presence of fixations and saccades

Abstract

Ternary eye movement classification, which separates fixations, saccades, and smooth pursuit from the raw eye positional data, is extremely challenging. This article develops new and modifies existing eye-tracking algorithms for the purpose of conducting meaningful ternary classification. To this end, a set of qualitative and quantitative behavior scores is introduced to facilitate the assessment of classification performance and to provide means for automated threshold selection. Experimental evaluation of the proposed methods is conducted using eye movement records obtained from 11 subjects at 1000 Hz in response to a step-ramp stimulus eliciting fixations, saccades, and smooth pursuits. Results indicate that a simple hybrid method that incorporates velocity and dispersion thresholding allows producing robust classification performance. It is concluded that behavior scores are able to aid automated threshold selection for the algorithms capable of successful classification.

One algorithm to rule them all? An evaluation and discussion of ten eye movement event-detection algorithms

“…Although this is a simple task to fix, it affects the original algorithm and is thus no longer objectively evaluated by us. Finally, for practical reasons, we limited ourselves to ten algorithms which furthermore should have no support for smooth-pursuit identification, which complicates matters further and is beyond the scope of this paper (but see Komogortsev & Karpov, 2012, for an evaluation of these algorithms). A search for a set of ten algorithms that fulfilled these criteria produced the algorithms that are described in the following paragraphs (see Table 1).…”

Detection of fixations and smooth pursuit movements in high-speed eye-tracking data

“…Since the signal characteristics of fixations and smooth pursuit movements are overlapping [9], classification of fixations in the presence of smooth pursuit movements is a difficult task [5,10]. The task is also different depending on whether the algorithm is intended for analysis of data recorded with a high or low sampling frequency, and for real-time or offline processing.…”

“…In [10], three algorithms for detection of fixations, saccades, and smooth pursuit movements were evaluated: a velocity based algorithm with two velocity thresholds (I-VVT), a velocity and movement pattern based algorithm (I-VMP), and a velocity and dispersion based algorithm (I-VDT). All algorithms were evaluated with data recorded using the EyeLink 1000 from SR Research.…”

Fixation identification for low-sample-rate mobile eye trackers

“…The linear kernel is robust to overfitting and gives better speed than a non-linear kernel. We compare our method to i) velocity-based threshold (I-VT) [10], ii) hidden Markov model identification [16], and iii) the EyeMMV [11].…”