Driver's turning intent recognition model based on brain activation and contextual information

Trende, Alexander; Unni, Anirudh; Jablonski, Mischa; Biebl, Bianca; Lüdtke, Andreas; Fränzle, Martin; Rieger, Jochem W. · 2022 · Crossref

DOI: 10.3389/fnrgo.2022.956863

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-critical problem of driver turning intent recognition at unsignalized intersections, where human error often leads to accidents. While existing models rely primarily on contextual information like gap size and waiting time, this research investigates whether integrating brain activation data improves prediction accuracy. The authors hypothesize that neurophysiological measures provide a more direct indicator of decision-making, potentially enhancing the performance of advanced driver assistance systems. The researchers conducted a driving simulator study with 13 participants performing left-turn maneuvers through oncoming traffic. Brain activation was measured using high-density functional near-infrared spectroscopy (fNIRS), capturing whole-head data at approximately 2 Hz. Contextual variables included gap size and waiting time. Three deep neural network models were trained to classify turning intent ("turn" vs. "no turn"): one using only contextual data, one using only fNIRS data, and a combined model using both. The fNIRS data underwent preprocessing to remove physiological artifacts, followed by principal component analysis to reduce dimensionality. Model performance was evaluated using five-fold cross-validation, and feature importance was analyzed using SHapley Additive exPlanations (SHAP). The results demonstrated that the combined model significantly outperformed the individual models, achieving a median accuracy of 91.9%, compared to 83.8% for the context-only model and 83.1% for the fNIRS-only model. Crucially, the combined model reduced false negatives—instances where the driver intended to turn but the model predicted otherwise—to 2.5%, whereas the single-modality models misclassified approximately 10% of these safety-critical events. SHAP analysis revealed that gap size was the most influential feature, followed by specific fNIRS principal components. Neuroimaging results indicated increased activation differences during the decision phase in the left motor cortices (primary, premotor, and supplementary motor areas) and the left middle frontal gyrus, associated with executive functions and action planning. The study concludes that integrating neurophysiological data with contextual information significantly enhances the recognition of driver turning intent. By reducing false negatives, this multimodal approach offers a more reliable method for predicting driver behavior, which could improve the effectiveness of warning systems and risk-mitigation strategies in autonomous and semi-autonomous vehicles. The findings support the use of fNIRS as a viable, portable tool for real-time neuroergonomic applications in driving safety.

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.

• signaling behavior
• anticipation
• mental model of traffic
• situational awareness

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: computational model, theory or model