Improving Driver Situation Awareness Prediction using Human Visual Sensory and Memory Mechanism

Zhu, Haibei; Misu, Teruhisa; Martin, Sujitha; Wu, Xingwei; Akash, Kumar · 2021 · arXiv

archive: archived pipeline: cataloged verified

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

Summary

This paper addresses the challenge of improving Advanced Driver Assistance Systems (ADAS) by accurately predicting driver situation awareness (SA). Current ADAS warnings often prove redundant because they rely solely on traffic conditions and vehicle dynamics, ignoring whether the driver is already aware of hazards. To solve this, the authors propose a predictive model that integrates driver gaze behavior with object properties and human cognitive mechanisms, specifically visual sensory characteristics and short-term memory limits. The study utilizes data from a driving simulation experiment involving 44 participants who navigated predefined routes and answered SA questions using the Situation Awareness Global Assessment Technique (SAGAT) at eight specific pauses. The dataset comprises 1,232 data points regarding awareness of 28 target objects. The authors developed features across four categories: gaze point-based metrics (distance to objects), human visual sensory-dependent features (accounting for foveal and peripheral vision ranges), object spatial and property-based features (such as contrast, movement, and relevance to the ego-vehicle), and memory-based features. The memory component employs a two-classifier structure that ranks objects by awareness probability and applies a re-ranking function based on Miller’s Law (the "magical number seven") to approximate human short-term memory capacity. Experimental evaluation using leave-one-pause-out cross-validation demonstrates that the proposed methods significantly outperform three baselines: a rule-based fixation model, a learning-based fixation model, and a model using only gaze and spatial features. The baseline models achieved accuracies between 54.9% and 64.4%. Incorporating object property features improved accuracy to 69.9%, while adding visual sensory features raised it to 71.5%. The final model, which included the memory-oriented re-ranking strategy, achieved the highest accuracy of 72.4% and an Area Under the Curve (AUC) of 0.762. Analysis of Principal Component Analysis (PCA) weights revealed that object saliency features and peripheral vision tracking behaviors were critical contributors to prediction performance, whereas raw spatial features were less significant. The significance of this work lies in its demonstration that modeling human visual and memory mechanisms substantially enhances SA prediction accuracy compared to gaze-only approaches. By achieving over 70% accuracy, the model provides a viable foundation for real-time SA estimation. This capability enables future ADAS implementations to issue warnings only when drivers are unaware of hazards, thereby reducing redundant alerts and improving the collaboration between human drivers and intelligent vehicle systems. The authors note limitations regarding the simulator environment and data size, suggesting future work should validate these findings in real-world driving conditions.

Key finding

Augmenting gaze-based driver SA prediction with foveal/peripheral vision features and short-term-memory adjustments raises accuracy above 70%, outperforming the gaze-only baseline.

Methodology

lab_experiment

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. Discovered via discover_arxiv on 2026-05-04 (4 acquisition events logged).

Summary generated by qwen3.6-27b-prismaquant on 2026-06-10; verification: verified.

Topics

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

• situational awareness
• gaze based attention detection
• anticipation
• peripheral attention
• useful field of view
• attention allocation

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
• Theoretical Contribution: theory or model, computational model