Real-Time Driver Alertness and Assistance System

Abstract

This research presents an integrated safety monitoring system designed to enhance road safety by analyzing driver behaviour and vehicle interactions in real-time. The system utilizes a multi- layered deep learning architecture to detect potential hazards and evaluate driving performance. For ex ternal environment monitoring, a YOLOv8 object detection model is employed to identify and track various vehicle classes, including cars, bikes, buses, and trucks. To specifically evaluate overtaking operations, we developed an Advanced Overtaking LSTM model. This bidirectional LSTM network processes temporal feature sequences including horizontal motion and area change ratios to classify operations as “Safe” or “Rash” with high precision. In addition to external monitoring, the system incorporates a secondary YOLOv11-based model trained on specialized datasets to monitor the internal driver state. This model detects critical signs of fatigue and distraction, such as drowsiness and yawning. The core innovation lies in the seamless fusion of these disparate data streams into a unified processing pipeline. By correlating external operation risks with the internal physiological state of the driver, the system calculates a dynamic Driver Score (0–100). This score categorizes behaviour into Safe, Moderate, or Aggressive profiles, allowing for highly personalized safety interventions. To ensure practical utility, the entire backend is integrated into a mobile application developed using the React Native framework, enabling cross-platform accessibility and low-latency feedback. This application provides real-time feedback and auditory alerts, ensuring that high-level safety monitoring is accessible to any driver with a smartphone. Experimental evaluations indicate that the system maintains high accuracy in complex traffic conditions, significantly reducing the gap between advanced driver assistance systems (ADAS) and everyday mobile-based safety tools.

Introduction

The text proposes a low-cost, smartphone-based driver safety system that uses deep learning to reduce road accidents caused mainly by human errors such as fatigue and risky overtaking.

The system combines two main modules: an external environment analyzer and an internal driver monitor. The external module uses YOLO-based object detection with DeepSORT tracking and an LSTM model to analyze vehicle movement and classify overtaking behavior as safe or rash. The internal module uses a YOLO11-based model to detect driver states such as drowsiness, yawning, and eye closure in real time.

Both modules are fused into a Driver Scoring system, which continuously updates a safety score based on risky driving actions and fatigue levels. This score categorizes drivers into safe, moderate, or aggressive risk levels and triggers real-time alerts when necessary.

The system is deployed through a React Native mobile application, turning a smartphone into a real-time driver assistance tool without requiring expensive vehicle hardware. It is trained on GPU infrastructure but optimized for mobile inference.

Conclusion

This research presents a holistic safety solution that integrates internal driver behaviour monitoring with external situational analysis into a unified, real-time processing pipeline. Utilizing YOLO11 for facial landmark detection, the system accurately identifies physiological signs of impairment such as drowsiness and yawning, triggering immediate visual interventions. Simultaneously, the external module employs YOLOv8 and a Bidirectional LSTM to track surrounding vehicles and classify maneuvers as \"Safe\" or \"Rash\" based on temporal motion dynamics. These dual-perspective data streams are fused into a dynamic Driver Scoring algorithm delivered via a low-latency React Native mobile application. Experimental validation demonstrates that the system maintains an 80.7% overall recall rate on consumer-grade hardware, providing a scalable and accessible tool to enhance road safety and reduce accidents caused by human error.

References

[1] Behavioural Detection: Gupta and Sharma (2019) emphasize using behavioural characteristics like eye-closure rates and blinking frequency as a non-intrusive method for fatigue detection. [2] Facial Landmark Analysis: Sahayadhas, Sundaraj, and Murugappan (2013) demonstrate that facial landmark analysis effectively identifies yawning and prolonged eye closure. [3] Deep Learning Evolution: Ramzan et al. (2019) document the transition in driver drowsiness detection from traditional image processing to Convolutional Neural Networks (CNNs). [4] Real-time Performance: Singh and Kumar (2020) explore the essential trade-offs between accuracy and processing speed required for a real-time machine learning approach. [5] Vehicle Dynamics: Zhang, Wang, and Li (2021) investigate how analyzing vehicle motion data and OBD.

Copyright

Copyright © 2026 Ms. Priya Meshram , Karan Tulswani, Karan Parwani, Mahima Singh, Kamlesh Thote, Ayush Karmani, Koyal Rahangadale. This is an open access article distributed under the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.