BEHOLD: An Extensible Eye-Tracking Infrastructure Supporting Multimodal, Multi-Device Interaction Evaluation

Abstract

Interactive systems increasingly span multiple devices, modalities, and physical spaces, which makes interaction evaluation harder than in single-device settings. Traditional post-task methods (e.g., questionnaires and logs) often miss when and why problems occur. Eye tracking can complement these approaches by continuously capturing visual attention, helping reveal what users notice or overlook and how effort is distributed across interfaces. However, deploying eye tracking in such ecosystems raises challenges: different contexts require different trackers (wearables, environmental cameras, or display-mounted sensors); existing tools are often tied to specific hardware and may lack evaluation-oriented analysis; and synchronising gaze with other contextual and system-level data typically requires multiple disconnected components. This work presents the first stage of BEHOLD (Biometric Eye-tracking Hub for Observation, Logging, and Data), a plugin-based proof of concept that separates device-specific logic from a unified processing and analysis pipeline. Each eye tracker is supported via an individual plugin for parsing and validation, while the framework treats gaze streams consistently regardless of source. For storage, BEHOLD combines TimescaleDB for high-frequency gaze samples, PostgreSQL for session metadata, and MinIO for recordings and exports. We demonstrate the approach by integrating data from Tobii Pro Glasses, showing how the plugin architecture accommodates device-specific requirements while enabling a coherent workflow from acquisition to analysis.

Keywords: Eye Tracking, Interactive Ecosystems, Hardware-Agnostic Framework, Multimodal Evaluation

DOI: 10.54941/ahfe1007266