Real-Time Driver Alertness and Assistance System

Meshram, Ms. Priya · 2026 · Crossref

DOI: 10.22214/ijraset.2026.80062

archive: archived pipeline: cataloged verified

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

Summary

This paper addresses the critical issue of road safety, specifically targeting accidents caused by human error such as driver fatigue and aggressive driving behaviors like risky overtaking. The authors identify a significant gap in current technology: while Advanced Driver Assistance Systems (ADAS) exist in luxury vehicles, they are inaccessible to the majority of drivers due to high costs and specialized hardware requirements. To bridge this disparity, the research proposes a low-cost, real-time safety monitoring system that leverages ubiquitous smartphone technology and deep learning to provide comprehensive driver alertness and assistance. The system employs a multi-layered architecture integrating external environment monitoring and internal driver state analysis. For external monitoring, a YOLOv8 object detection model identifies and tracks vehicles (cars, bikes, buses, trucks) using the DeepSort algorithm. An Advanced Overtaking LSTM network, featuring 192 hidden units and a 40% dropout rate, analyzes temporal sequences of horizontal motion and area change ratios to classify overtaking maneuvers as “Safe” or “Rash.” For internal monitoring, a YOLOv11 model detects physiological signs of impairment, including drowsiness, yawning, and eye closure. These data streams are fused into a dynamic Driver Score (0–100), which categorizes driving behavior into Safe, Moderate, or Aggressive profiles. The entire backend is deployed via a React Native mobile application, enabling cross-platform accessibility and real-time auditory and visual alerts. Experimental evaluations demonstrate the system’s effectiveness and practicality. The internal YOLOv11 model achieved an overall Recall of 0.807 on a validation set of 1,881 images, with the “Drowsy eye” class reaching a mean Average Precision (mAP@50) of 0.841. The system prioritized high recall to minimize false negatives in safety-critical scenarios. The LSTM model successfully distinguished between safe and rash overtaking operations by analyzing temporal dynamics. Crucially, the mobile implementation maintained an average frame rate of 22 FPS with an end-to-end latency of approximately 95ms on consumer-grade smartphone hardware, meeting the threshold for real-time safety intervention. The significance of this work lies in its ability to democratize advanced safety monitoring. By achieving high accuracy and low latency on standard smartphones, the system eliminates the need for expensive embedded GPU clusters or specialized OBD-II hardware. This approach significantly reduces the technological gap in road safety tools, making sophisticated ADAS features accessible to everyday drivers. The study concludes that integrating behavioral and situational analysis into a unified, mobile-based pipeline offers a scalable solution to reduce accidents caused by human error, although future work is needed to address performance in low-light conditions.

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.

• drowsiness detection algorithms
• distraction detection algorithms
• dms validation
• in vehicle coaching
• gaze based attention detection

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