Sleep Detection System for Trucks: A Real-Time Multi-Modal Data-Driven AI Framework for Driver Fatigue Monitoring

Verma, Paras · 2025 · Crossref

DOI: 10.22214/ijraset.2025.68705

archive: archived pipeline: cataloged

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

Summary

This paper addresses the critical safety issue of driver fatigue in commercial trucking, a leading cause of road accidents that current single-modality systems fail to detect reliably due to environmental noise and individual variability. The study proposes a real-time, multi-modal Artificial Intelligence (AI) framework designed specifically for heavy vehicle environments to monitor driver drowsiness and trigger proactive alerts. The system integrates three distinct input modalities: behavioral signals via Eye Aspect Ratio (EAR) calculated from facial landmarks, physiological indicators through Heart Rate Variability (HRV) metrics such as RMSSD and LF/HF ratios, and vehicular dynamics using Steering Entropy (SE) derived from CAN-bus telemetry. These data streams are processed by a hybrid deep learning architecture that combines Convolutional Neural Networks (CNN) for spatial feature extraction with Long Short-Term Memory (LSTM) networks to capture temporal fatigue trends. The model was trained on the NTHU Drowsy Driver Dataset augmented with synthetic driving footage. Experimental evaluation was conducted in a simulated truck cabin environment using an NVIDIA Jetson Nano, a Logitech C920 IR camera, a MAX30102 HRV sensor, and an MCP2515 CAN controller. The testing protocol included five specific scenarios ranging from normal driving to simulated microsleeps, erratic steering, and physiological fatigue states. Results indicate that the hybrid model achieved a classification accuracy of 94.1%, with precision of 92.6%, recall of 93.5%, and an F1-score of 93.0% on a test set of 1,000 labeled sequences. The system demonstrated an average detection latency of 124 milliseconds, meeting the target for real-time responsiveness. Feature contribution analysis revealed that EAR was the most significant predictor, followed by HRV and Steering Entropy, confirming the complementary nature of the fused inputs. The multi-modal approach outperformed single-modality methods, particularly in low-light conditions or when visual data was obstructed, as HRV and steering data provided redundant detection channels. Additionally, the system maintained high accuracy during circadian rhythm dips in the early morning and post-lunch hours. The significance of this research lies in its demonstration that a low-cost, edge-computing-based AI system can effectively monitor driver fatigue in commercial trucks, addressing a gap in current industry solutions which are often not ruggedized or optimized for long-haul operations. By achieving high accuracy with low latency and minimal intrusiveness, the framework supports United Nations Sustainable Development Goal 3.6, aiming to halve global road traffic deaths and injuries by 2030. The study concludes that multi-modal sensor fusion is essential for robust detection in noisy real-world environments, and the proposed architecture offers a scalable, privacy-compliant solution for commercial fleets.

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 qwen3.8-27b-gittensor on 2026-08-23; verification: pending re-verification.

Topics

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

• drowsiness detection algorithms
• truck driver fatigue
• dms validation
• distraction detection algorithms
• drowsiness
• microsleep

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).

• Empirical Findings: physiological data
• Methodological Resource: tool software, validation psychometrics