Adaptive Neural Embedded Systems for Real-Time Driver State Monitoring in Advanced Driver Assistance Systems

Alagar Raja Govindasamy · 2025 · Crossref

DOI: 10.52783/jisem.v10i63s.13851

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 challenge of monitoring driver cognitive and physical states during the transition from SAE Level 2 to Level 3 vehicle autonomy, where the driver’s role shifts from active controller to supervisor. The motivation stems from the high mortality and economic burden associated with driver fatigue and distraction, which contribute to an estimated 1,550 deaths and $12.5 billion in annual losses in the United States. The authors propose an adaptive neural embedded system that integrates multimodal sensor fusion with edge-deployed artificial intelligence to provide real-time, personalized driver state monitoring, overcoming the latency, connectivity, and privacy limitations of cloud-dependent systems. The system architecture employs a hybrid neural framework combining Convolutional Neural Networks (CNNs) for spatial feature extraction and Long Short-Term Memory (LSTM) networks for temporal pattern recognition. This hybrid model processes facial images, physiological signals, and behavioral indicators to detect fatigue and distraction. Implementation occurs on embedded FPGA-System-on-Chip platforms, utilizing optimized models such as MobileNetV2 and EfficientNet-B0 to achieve sub-second inference latency. The system utilizes multimodal sensor fusion, integrating infrared eye-tracking, head pose estimation, and steering behavior monitoring. Key metrics include Eye Aspect Ratio (EAR), Mouth Aspect Ratio (MAR), and Percentage of Eye Closure (PERCLOS). The architecture supports baseline calibration for driver-specific thresholds and continuous model refinement through incremental learning, allowing the system to adapt to changing behavioral patterns over time. Experimental results indicate that the proposed system achieves high classification accuracy across diverse operational environments. Specific implementations using ResNet-50 backbones with attention mechanisms reached training accuracies of 99.87% and validation accuracies of 97.86%. Embedded deployments using MobileNetV2 achieved 96% classification accuracy with inference times as low as 31 milliseconds per frame. Transfer learning techniques, specifically fine-tuning InceptionV3 architectures on domain-specific drowsiness datasets, yielded a classification accuracy of 99.25%. The system maintains deterministic real-time performance, with end-to-end processing times below 35 milliseconds per frame, enabling graduated intervention strategies ranging from sensory alerts to active vehicle control inputs like lane-centering assistance. The significance of this work lies in its demonstration of a robust, privacy-preserving, and low-latency solution for driver monitoring in advanced driver assistance systems. By moving processing to the vehicle edge, the system ensures reliable performance independent of network connectivity, which is essential for safety-critical applications. The integration of adaptive learning and personalized calibration addresses the variability in individual physiological baselines, reducing false positives and enhancing the reliability of drowsiness detection. This approach provides a scalable framework for integrating intelligent safety interventions into increasingly autonomous vehicular environments, directly addressing the causal factors of driver error in traffic fatalities.

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
• distraction detection algorithms
• drowsiness
• gaze based attention detection
• dms validation
• 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
• Theoretical Contribution: computational model