Decoding Human Behavior through Oculomotor Feature Extraction

24 concepts23 papers1 proposal

Raw eye-tracking data is deceptively simple: a high-frequency stream of $(x, y)$ coordinates representing where a person is looking. On its own, this data is noisy and lacks semantic meaning. To transform these coordinates into insights about a user's fatigue, expertise, or personality, researchers use oculomotor feature extraction. This process distills raw gaze traces into structured descriptors that capture the underlying mechanics of human vision and cognition.

The Fundamental Partition: Fixations and Saccades

The first step in making sense of gaze data is recognizing that the human eye does not move like a smooth cursor. Instead, it alternates between rapid jumps and stable pauses. As Gao et al. and Kasneci et al. describe, researchers typically use Velocity-Threshold Identification (I-VT) to segment raw data.

The intuition is simple: when the velocity of the gaze exceeds a certain threshold, the eye is in a saccade—a rapid transition between targets during which visual sensitivity is suppressed. When the velocity drops below that threshold, the eye is in a fixation, the period where the brain actually extracts information. By calculating these kinematic features, algorithms like SaFiDe, proposed by Madariaga et al., can objectively distinguish between periods of stable information processing and the motor control required to shift attention.

Engineering High-Dimensional Descriptors

Once gaze is segmented into events, researchers engineer higher-level features to capture behavioral "signatures." These features generally fall into three categories:

  1. Statistical Aggregates: Metrics like mean fixation duration, saccade length, and transition frequencies. Gingerich et al. demonstrate that these aggregated metrics can predict task complexity and user performance in real-time.
  2. Temporal and Spatial Patterns: Beyond simple averages, the sequence of movements matters. Dong et al. extracted 250 distinct features, including saccade direction and temporal sequences via sliding windows, to identify how users read maps. Similarly, Tseng et al. utilize gaze entropy to quantify the transition from focused tracking to erratic, fatigued movements.
  3. Micro-patterns: Raue et al. transform raw data into discrete patterns like "regressions" (looking back at previous text) or "sweep returns" to classify reading behaviors like skimming versus scanning.

Multimodal Fusion and Context

Oculomotor features become even more powerful when fused with other physiological or environmental data. The eye is not just a pointer; it is a biological organ. Kasneci et al. show that combining 2D pupil centers with 3D optical vectors allows for more robust classification of complex movements like smooth pursuits.

Furthermore, the context of the gaze is vital. Shiga et al. combined gaze motion with visual features of the scene to recognize daily activities, while Dreisig et al. integrated eye blink features (like PERCLOS) with head nodding to detect driver drowsiness. This multi-modal approach ensures that the system captures the full physiological state of the user.

From Features to Behavioral Inference

Why go through this extensive feature engineering? The goal is to build a bridge between physiological "leaks"—subtle, involuntary eye movements—and internal cognitive states.

  • Personality and Preference: Hoppe et al. found that eye movement signatures are consistent enough to predict "Big Five" personality traits during everyday errands. Sugano et al. showed that saccade counts and fixation frequencies are better predictors of image preference than standard image metadata.
  • Expertise and Skill: In highly dynamic environments, the way an expert searches for information differs from a novice. Hosp et al. used sequences of fixation image patches to classify the expertise of soccer goalkeepers, capturing spatial and temporal nuances that raw coordinates miss.
  • Health and Fatigue: Oculomotor features serve as digital biomarkers. Zhai et al. used scanpath metrics to objectively measure eye fatigue in VR, while Mehringer et al. used gaze velocity and search patterns to identify neurological indicators of Post-COVID syndrome.

By distilling raw gaze into these precise oculomotor features, researchers can move beyond simple heatmaps to create adaptive systems that understand the user's intent, state, and identity.

Key Foundational Work

Constructing Models of User and Task Characteristics from Eye Gaze Data for User-Adaptive Information Highlighting (2015) — contributed 1 concept elements to this direction.
Daily activity recognition combining gaze motion and visual features (2014) — contributed 1 concept elements to this direction.
Image preference estimation with a data-driven approach: A comparative study between gaze and image features (2014) — contributed 1 concept elements to this direction.
History & citationVersion 1 · 2026-10-04

Current · Permanent link

View frozen version 1

Cite this direction

Latest version (a living page):

@misc{lacuna_oculomotor_feature_extraction_2a467350,
  title = {oculomotor feature extraction},
  author = {{Lacuna}},
  howpublished = {Lacuna research direction},
  year = {2026},
  url = {https://lacuna.tiptreesystems.com/direction/oculomotor-feature-extraction/rdr_2a4673500b234e01b0461015bcd5527f},
  note = {Latest version of a living research direction; cite a version for a fixed text. Accessed 2026-11-21},
}

This exact version (version 1, frozen text):

@misc{lacuna_oculomotor_feature_extraction_2a467350_v1,
  title = {oculomotor feature extraction},
  author = {{Lacuna}},
  howpublished = {Lacuna research direction},
  year = {2026},
  url = {https://lacuna.tiptreesystems.com/direction/oculomotor-feature-extraction/rdr_2a4673500b234e01b0461015bcd5527f/v/1},
  note = {Version 1, frozen 2026-10-04. Accessed 2026-11-21},
}
About

These directions overlap with this one through shared papers.

See something wrong? Report a mistake