Visual Fixations Duration as an Indicator of Skill Level in eSports

Boris B. Velichkovsky, Nikita Khromov, Alexander Korotin, Evgeny Burnaev, Andrey Somov

TL;DR

Researchers have discovered that the way an eSports player’s eyes move is a "dead giveaway" for their skill level. By analyzing the duration of visual fixations in Counter-Strike: Global Offensive (CS:GO), the study found that professional players exhibit a unique "bimodal" distribution of gaze—alternating between rapid spatial scanning and deep conscious processing—whereas novices stick to a much more uniform, middle-of-the-road rhythm.

What Problem Does This Paper Address?

In the high-stakes world of eSports, success depends on a complex mix of cognitive, motor, and visual skills. While we know pros are "better," quantifying why from a Human-Computer Interaction (HCI) perspective is difficult.

The authors address a specific gap: Can we use eye-tracking data—specifically the duration of fixations (when the eye stays still to take in information)—as an objective metric for expertise? Understanding these patterns isn't just about scouting talent; it’s about designing "adaptive interfaces" that can react to a user’s skill level or cognitive load in real-time.

Methods

The researchers built an in-house IoT platform to sync data from multiple sources:

  • Hardware: A Tobii EyeX eye-tracker (sampling at 30 Hz), a Garmin Heart Rate Monitor, and a standard gaming PC.
  • Software: A custom script captured the gaze stream into CSV files, while an in-game logger tracked match statistics.
  • Fixation Extraction: They used the Mould algorithm within the gazepath R package. This is a key technical detail because the algorithm adapts its detection thresholds on a subject-by-subject basis, accounting for individual physiological differences.
  • Analysis: The team employed three layers of statistical analysis:
  • Descriptive Statistics: Comparing means, medians, and standard deviations.
  • Vincentile Analysis: Breaking the duration distributions into five equal "bins" to see how the tails of the data behave.
  • Kernel Density Estimation (KDE): A non-parametric way to estimate the probability density function, helping them visualize whether the distributions were unimodal (one peak) or bimodal (two peaks).

Datasets

The study utilized a primary dataset collected from 21 participants divided into three groups based on their CS:GO experience:

  • Low Skill (N=10): Amateurs with less than 700 hours of playtime.
  • High Skill (N=7): Experienced amateurs with over 700 hours.
  • Pro (N=4): Professional athletes from the "Monolith" team (Russia) with over 10,000 hours of experience (averaging 4+ hours of play per day).

The data was gathered during a specific "Retake" scenario in CS:GO, involving 12 rounds of 40 seconds each.

Key Insights and Ideas

  • The Bimodal Shift: The most striking finding is that as skill increases, the distribution of fixation durations shifts from a single peak to two distinct peaks.
  • Ambient vs. Focal Processing: The authors map these two peaks to established psychological concepts:
  • Ambient Fixations (~100ms): Rapid, automatic spatial processing (the "Where" system). Pros use these to maintain situational awareness without "thinking" about it.
  • Focal Fixations (>300-500ms): Conscious, deliberate identification of objects (the "What" system).
  • Novices are "Stuck in the Middle": Low-skill players have a very uniform distribution. They don't show the ultra-fast ambient scans or the very long, concentrated focal fixations seen in pros. Their visual strategy is less specialized.
  • Variability is a Virtue: In many fields, "consistency" is the hallmark of a pro. In visual processing for eSports, the opposite is true: higher skill leads to wider and more variable fixation distributions.

Key Results

  • Standard Deviation (SD): There was a "systematic increase" in SD with skill level ($p < 0.001$). Low-skill SD was 122ms, while Pro SD jumped to 223ms.
  • Minimum/Maximum Durations: Pros had significantly lower minimum fixation durations (~55ms) and significantly higher maximums (~1229ms) compared to novices (min: 117ms, max: 800ms).
  • The "Aha" Moment in KDE: The Kernel Density plots (Figure 3 in the paper) clearly show the Pro group developing a second peak around the 100ms mark, which is almost entirely absent in the low-skill group.
  • Vincentile Interaction: A 3x5 rm-ANOVA showed a highly significant interaction ($F(8,72)=10.3, p<0.001$), confirming that the "shape" of the gaze distribution changes fundamentally as a player levels up.

Limitations and Future Work

  • Sample Size: The study is small, particularly the "Pro" group (N=4). While the results are statistically significant, a larger cohort would strengthen the findings.
  • Game Specificity: The authors acknowledge that these results might not generalize to other genres (like RTS or MOBAs) where the visual demands differ from a first-person shooter.
  • Hardware Constraints: The 30 Hz sampling rate of the Tobii EyeX is relatively low for capturing the finest nuances of eye movement; higher-frequency trackers might reveal even more detail.
  • Future Application: The authors suggest using this data to create Adaptive UIs. For example, if a system detects a player is only using "ambient" fixations on a specific UI element, it could increase that element's saliency (size/color) to force "focal" attention.

Summary updated

Abstract

Using highly interactive systems like computer games requires a lot of visual activity and eye movements. Eye movements are best characterized by visual fixation – periods of time when the eyes stay relatively still over an object. We analyzed the distributions of fixation duration of professional athletes, amateur and newbie players. We show that the analysis of fixation durations can be used to deduce the skill level in computer game players. Highly skilled gaming performance is characterized by more variability in fixation durations and by bimodal fixation duration distributions suggesting the presence of two fixation types in high skill gamers. These fixation types were identified as ambient (automatic spatial processing) and focal (conscious visual processing). The analysis of computer gamers’ skill level via the analysis of fixation durations may be used in developing adaptive interfaces and in interface design.
BibTeX
@inproceedings{velichkovsky2019visual,
  author = {Boris B. Velichkovsky and Nikita Khromov and Alexander Korotin and Evgeny Burnaev and Andrey Somov},
  title = {{Visual Fixations Duration as an Indicator of Skill Level in eSports}},
  booktitle = {INTERACT},
  year = {2019},
  url = {https://openreview.net/forum?id=nrDIpGQpnD2},
}

Loading related content...

Figure