[Paper Review] Eye Movement Feature Classification for Soccer Goalkeeper Expertise Identification in Virtual Reality

Benedikt Hosp, Florian Schultz+2 more|arXiv (Cornell University)|Sep 23, 2020

Sport Psychology and Performance38 references4 citations

TL;DR

This study proposes a machine learning approach to classify soccer goalkeeper expertise using eye movement features captured during virtual reality (VR) simulations of real-game scenarios. By analyzing gaze behavior from 35 goalkeepers across expert, intermediate, and novice levels, the method achieves promising classification accuracy, demonstrating that eye-tracking data in immersive VR can objectively assess perceptual-cognitive expertise and inform adaptive training systems.

ABSTRACT

The latest research in expertise assessment of soccer players has affirmed the importance of perceptual skills (especially for decision making) by focusing either on high experimental control or on a realistic presentation. To assess the perceptual skills of athletes in an optimized manner, we captured omnidirectional in-field scenes and showed these to 12 expert, 10 intermediate and 13 novice soccer goalkeepers on virtual reality glasses. All scenes were shown from the same natural goalkeeper perspective and ended after the return pass to the goalkeeper. Based on their gaze behavior we classified their expertise with common machine learning techniques. This pilot study shows promising results for objective classification of goalkeepers expertise based on their gaze behaviour and provided valuable insight to inform the design of training systems to enhance perceptual skills of athletes.

Motivation & Objective

• To develop an objective, data-driven method for assessing soccer goalkeeper expertise using eye movement features.
• To evaluate whether gaze behavior in immersive VR environments can differentiate between expert, intermediate, and novice goalkeepers.
• To explore the feasibility of using machine learning on eye-tracking data for real-time expertise classification in sports training.
• To inform the design of adaptive virtual reality training systems that respond to individual perceptual-cognitive strengths and weaknesses.
• To establish a foundation for future online diagnostic and personalized training systems using real-time gaze analysis.

Proposed method

• Captured 360° video stimuli from the natural goalkeeper perspective during real match situations to ensure ecological validity.
• Presented these stimuli to 35 participants (12 experts, 10 intermediates, 13 novices) using consumer-grade VR headsets with integrated eye trackers.
• Extracted gaze behavior features such as fixation duration, saccade frequency, and spatial distribution across key game areas.
• Applied standard supervised machine learning techniques (e.g., SVM, random forests) to classify participants into expertise levels based on gaze features.
• Used real-time data processing pipelines with 250 Hz eye tracker sampling, enabling online feature computation and model inference.
• Designed a multi-threaded system to compute higher-level features (e.g., standard deviations) post-trial and feed them into a pre-trained model for real-time classification.

Experimental results

Research questions

• RQ1 Can eye movement features in a VR environment reliably differentiate between expert, intermediate, and novice soccer goalkeepers?
• RQ2 To what extent does gaze behavior in immersive VR reflect real-world perceptual-cognitive expertise in goalkeepers?
• RQ3 Can machine learning models trained on gaze data achieve accurate and objective classification of goalkeeper expertise?
• RQ4 How can gaze-based expertise detection be integrated into real-time, adaptive training systems?
• RQ5 What are the limitations of current eye-tracking and feature extraction methods in capturing subtle perceptual differences?

Key findings

• The study achieved promising classification accuracy in distinguishing between expert, intermediate, and novice goalkeepers based solely on gaze behavior in VR.
• Experts exhibited more efficient gaze patterns, such as faster cue detection and better spatial prioritization, compared to novices.
• The use of photo-realistic 360° VR stimuli preserved natural gaze behavior, enhancing the ecological validity of the data.
• Machine learning models demonstrated potential for real-time expertise classification using online eye-tracking data at 250 Hz.
• The results support the feasibility of using gaze-based diagnostics to personalize training by identifying individual perceptual weaknesses.
• Future work is planned to expand the dataset, refine classification granularity, and integrate the system into real-time adaptive VR and AR training platforms.