The application of an RGB-D camera for monitoring the allocation of visual attention among high-speed train drivers

Shen, Weiyi; Guo, Beiyuan · 2025 · Crossref

DOI: 10.54941/ahfe1006229

archive: archived pipeline: cataloged verified

Get this paper ↗ (DOI — opens at the source; we link to it, we don't host it)

Summary

This study addresses the safety risks associated with high-speed train drivers experiencing a decline in situation awareness (SA) while performing supervisory tasks in highly automated environments. As Automatic Train Operation (ATO) systems assume more driving responsibilities, drivers transition to monitoring roles, which can lead to reduced attention and delayed emergency responses. To mitigate this, the authors propose a non-contact monitoring method using an RGB-D camera to track visual attention allocation, positing that changes in visual focus correlate with SA levels. The methodology involves capturing 3D facial data to estimate head pose and eye gaze, which are then fused to determine the driver’s actual gaze direction. Head pose is calculated using 68 facial landmarks, specifically focusing on the forehead and cheeks to minimize interference from hair or glasses. Eye gaze is derived by locating the iris center and calculating a vector from the eye center, estimated using anatomical depth constraints. These two data streams are weighted at 0.7 for head pose and 0.3 for eye gaze to account for the broader field of view provided by head movements. The resulting gaze vectors are mapped onto Areas of Interest (AOIs) within the driver’s cab, including the forward railway, the Automatic Train Protection (ATP) screen, and the Train Control and Management System (TCMS) screen. Experiments were conducted with eight subjects using a high-fidelity CR400BF train simulator. Participants performed a 55-minute supervisory task in ATO mode, during which their SA was assessed via Situation Awareness Global Assessment Technique (SAGAT) queries triggered by random interruptions. Data analysis revealed that SA levels decreased significantly by the final query (mean score 0.53) compared to earlier queries (mean score 0.82), attributed to fatigue. Visual attention analysis showed that while the forward railway remained the primary focus, drivers with high SA allocated significantly more attention to the ATP screen (AOI-2). Specifically, fixation density and duration on the ATP screen were higher during high SA states, whereas low SA states were characterized by a narrowed field of view and neglect of the Driver Machine Interface (DMI). The findings demonstrate that RGB-D cameras effectively monitor visual attention shifts linked to SA degradation. The study concludes that maintaining high SA requires active engagement with the ATP screen, suggesting that optimizing information display on this interface could enhance driver supervision efficiency and operational safety. This approach provides a technical foundation for future DMI design and real-time monitoring systems in autonomous rail operations.

Provenance

The full processing record for this entry. Every stage of this paper's journey through the pipeline is logged — what ran, with which tool and model, how many attempts it took, and when it last completed.

Summary generated by qwen3.6-27b-nvidia on 2026-08-10; verification: verified.

Topics

Ranked by relevance to this paper. Hover a topic for its definition.

• gaze based attention detection
• attention allocation
• useful field of view
• peripheral attention
• distraction detection algorithms
• looked but failed to see

Information type

What kind of knowledge this paper contributes, grouped by family — independent of topic (what it is about) and method (how it was studied).

• Methodological Resource: tool software, measurement protocol