Pulsedrive-AI-based Driver Health Monitoring System for Smart & Heavy Vehicles

doi:https://doi.org/10.5281/zenodo.19314440

Volume 15, Issue 03 (March 2026)

Pulsedrive-AI-based Driver Health Monitoring System for Smart & Heavy Vehicles

DOI : https://doi.org/10.5281/zenodo.19314440

Download Full-Text PDF Cite this Publication

Open Access
Article Download / Views: 7
Authors : Dr. S. Sathish Kumar, Mr Anbazhagan. R, Mr. Jayakrishna. B, Mr. Arun Kumar S
Paper ID : IJERTV15IS030927
Volume & Issue : Volume 15, Issue 03 , March – 2026
Published (First Online): 29-03-2026
ISSN (Online) : 2278-0181
Publisher Name : IJERT
License: This work is licensed under a Creative Commons Attribution 4.0 International License

PDF Version

View

Text Only Version

Pulsedrive-AI-based Driver Health Monitoring System for Smart & Heavy Vehicles

Dr. S. Sathish Kumar, Mr Anbazhagan. R, Mr. Jayakrishna. B, Mr. Arun Kumar S

Scholars Dept of Artificial Intelligence and Data Science

Vel Tech High Tech Dr Rangarajan Dr Sakunthala Engineering College, Tamilnadu, Chennai, India

Abstract – Road accidents resulting from driver health issues have emerged as a critical safety concern, particularly among long-distance and public transport drivers. Sudden fatigue, oxygen deprivation, or cardiovascular irregularities often go undetected until they lead to serious incidents. To address this problem, the proposed systemPulseDrive: A Real-Time In- Cabin Health Monitoring Systemoffers an intelligent, non- intrusive solution for continuous physiological monitoring and early risk detection. The system integrates multiple biometric sensors, including the MAX30100, to measure pulse rate and blood oxygen saturation (SpO), along with temperature and fatigue detection modules for comprehensive health assessment. An ESP32 microcontroller acts as the central processing unit, aggregating sensor data and transmitting it to an AI-based analytics model for real-time anomaly detection. In the event of critical deviations, the system activates audio-visual alerts and automatically sends emergency notifications via the SIM800L GSM module to fleet operators or designated contacts. Additionally, all health data is logged for predictive analysis and long-term wellness tracking. By leveraging AI, IoT, and embedded system technologies, PulseDrive enhances road safety, supports proactive health management, and contributes to the development of smarter and safer transportation systems.

Keywords – Driver health monitoring, IoT, Artificial Intelligence, Embedded systems, Road safety, MAX30100, ESP32, GSM module, Fatigue detection.

I . INTRODUCTION

Background

In modern transportation, driver health and safety are critical, especially for those operating long-distance or commercial vehicles. Prolonged driving hours, irregular rest, and stress can cause fatigue or sudden medical issues, posing serious risks to all road users. While advancements in smart vehicle technologies emphasize vehicle performance and navigation, the drivers physiological well-being often remains overlooked. Addressing this gap forms the key motivation behind the PulseDrive project.
Problem Definition

Despite rapid progress in automotive systems, there is still no reliable solution for real-time monitoring of a drivers vital health parameters. Conditions such as hypoxia, irregular heart rate, or fatigue frequently go undetected

until critical situations arise. A real-time, non-invasive, in-cabin health monitoring system is therefore essential to identify these anomalies early and trigger preventive responses.

1.2 Objective

The primary goal of PulseDrive is to develop a real-time in- cabin health monitoring system that continuously tracks key physiological indicators such as heart rate, oxygen saturation (SpO), and fatigue levels. Using embedded AI analytics, the system detects irregularities and issues timely audio-visual or GSM-based alerts. Ultimately, PulseDrive aims to advance human-centric smart mobility by integrating driver wellness with intelligent transportation, ensuring safer and more responsive road environments.

II . SYSTEM SIGNIFICANCE AND PROBLEM CONTEXT

Significance of Driver Health Monitoring

Drivers are the backbone of the transportation network, ensuring safe and efficient mobility of people and goods. However, long working hours, stress, and irregular sleep patterns often lead to fatigue and health deterioration, increasing accident risks. Conventional safety systems focus mainly on vehicle parameterssuch as brakes or airbags while neglecting the drivers physiological state. PulseDrive addresses this gap by continuously tracking vital signs like pulse rate, oxygen saturation, and fatigue levels to ensure driver well-being and prevent health- related accidents.
Problem Definition

Despite advancements in automotive safety, the lack of real- time driver health monitoring remains a critical issue. Health conditions like hypoxia, elevated heart rate, or fatigue often go undetected until emergencies occur, endangering both drivers and passengers. Wearable devices provide limited solutions due to poor compliance and inconvenience. PulseDrive bridges this technological gap

by integrating non-invasive sensors, AI analytics, and IoT connectivity to continuously monitor, detect, and alert in case of abnormal health conditions, enhancing proactive road safety.
Relevance to AI in Automobiles

Artificial Intelligence (AI) is revolutionizing modern vehicles through automation and predictive safety systems.

However, most applications emphasise vehicle control rather than driver wellness. PulseDrive extends AIs scope by analyzing physiological data in real time using machine learning algorithms to detect fatigue, stress, or oxygen deprivation. It generates predictive alerts, adapts through continuous learning, and communicates with fleet management systems for timely intervention. By merging AI, IoT, and embedded sensing, PulseDrive exemplifies a human-centric approach to intelligent transportation, prioritizing both safety and health.

III. LITERATURE REVIEW

AI-Driven Driver Behaviour Assessment through Vehicle and Health Monitoring

Authors: Shumayla Yaqoob, Giacomo Morabito, Salvatore Cafiso, Giuseppina Pappalardo, and Ata Ullah Publication: IEEE

This study investigates the use of Artificial Intelligence (AI) to analyze driver behavior and physiological signals for improved road safety. It identifies a key research gap in linking driver health parameters with driving performance. Using image and signal processing, the system detects fatigue, distraction, and stress-induced anomalies in real time. The research introduces a taxonomy of detection

Keywords: Deep learning, predictive diagnostics, digital twin, AI-driven maintenance, automotive reliability.

3.3 A Secure and Intelligent Framework for Vehicle Health Monitoring

Authors: Md. Arafatur Rahman, Md. Abdur Rahim, Md. Mustafzur Rahman, Nour Moustafa, Imran Razzak, Tanvir Ahmad, and Mohammad N. Patwary

Publication: IEEE

This paper proposes a secure, IoE-based vehicle health monitoring framework using Multi-Layer Heterogeneous Networks (HetNet) and machine learning analytics. The system continuously collects and processes vehicle data to detect faults, alert drivers, and store diagnostic records for future use. Its big-data integration ensures scalability, while its intelligent analytics improve safety and maintenance efficiency. The proposed architecture paves the way for centralized, AI-enabled vehicular condition monitoring under Industry 4.0 standards.

Keywords: Vehicle health monitoring, IoE, HetNet, machine learning, predictive analytics, secure automotive systems.

IV . MATERIALS AND METHODS
Deep Learning for Predictive Vehicle Health Diagnostics

Author: Love David Adewale Publication: International Journal of Engineering Technology Research & Management (IJETRM)

This research explores deep learning methodssuch as CNNs, RNNs, and transformersfor predictive vehicle health monitoring and failure prevention. By analyzing sensor and telematics data, the models identify early fault indicators and predict mechanical or system malfunctions.

System Deployment and Validation
1. Arduino IDE: Used to program, compile, and upload embedded C++ code to the ESP32.
2. Edge Impulse Studio: A cloud-based TinyML platform utilized for:
  - Data collection and labeling
  - Signal preprocessing and feature extraction
  - Training and validating a Random Forest classifier
  - Deploying the optimized model as an Arduino- compatible library

Key Arduino Libraries:

Adafruit_MPU6050 IMU communication
SparkFun_MAX3010x Heart rate and SpO

sensing

Adafruit_GFX & Adafruit_SSD1306 Display handling
SD.h File operations for local data logging

4.3 Methodology

The methodology outlines the step-by-step development and integration of the PulseDrive system, ensuring efficient data flow and real-time response.

System Architecture and Data Flow

The operational workflow of PulseDrive follows a continuous data cycle, as illustrated in Figure IV.1 System Architecture Diagram (to be included in your report).
1. Data Acquisition: ESP32 retrieves sensor data
  
  (HR, SpO, motion) via the I²C interface.
2. Data Buffering: Sensor readings are grouped into 2-second time windows for stable inference.
3. AI Inference: The TinyML model analyzes buffered data and classifies the drivers state as Alert or Drowsy.
4. Action and Feedback: Upon detecting fatigue, the system activates the buzzer, RGB LED, and OLED warning message.
5. Data Logging: Sensor values and model outputs are stored on the SD card for future analysis.
TinyML Model Development

AI-based classification is achieved using Edge Impulse, enabling local intelligence without cloud dependency.
Data Collection:
1. Recorded sensor signals (HR, SpO, AccelX, AccelY, AccelZ) under two conditions Alert and Drowsy.
2. Data transmitted to Edge Impulse using the Data Forwarder utility.
Feature Extraction (Impulse Design):
1. Data segmented into 2-second windows.
2. Spectral Analysis applied to extract key statistical and frequency-domain features.
Model Training:
1. A Random Forest Classifier was trained using an 80:20 train-test split.
2. Validation results analyzed using confusion matrices to ensure high accuracy and low false detections.
Model Optimization and Deployment:
1. Quantized and compressed for microcontroller compatibility.
2. Exported as an Arduino library and executed using the run_classifier() function for real-time inference.
System Integration and Testing
1. Prototyping: Al modules connected on a breadboard for functionality testing.
2. Code Integration: Combined sensor data acquisition, OLED output, SD logging, and ML inference in a unified control loop.
3. Testing Scenarios: Simulated driving sessions conducted to evaluate model accuracy, alert response time, and power efficiency.
4. Validation: The system achieved reliable detection of fatigue states, demonstrating its potential for real- time deployment in commercial vehicles.

V. SYSTEM DESIGN AND IMPLEMENTATION

System Design and Implementation

This chapter presents the design and implementation details of PulseDrive an AI-Powered Driver Health and Fatigue Monitoring System. The system is developed to enhance the

safety of bus and heavy vehicle drivers by continuously monitoring their physiological parameters and detecting fatigue or abnormal health conditions in real time.

PulseDrive integrates embedded sensors, a microcontroller, and AI-based analytics to monitor vital parameters such as heart rate and oxygen saturation (SpO), detect fatigue, and generate timely alerts. This human- centric approach ensures proactive safety intervention, promoting both driver well-being and accident prevention.
1. System Overview
  
  The system is organized into three primary subsystems to ensure modularity, scalability, and efficiency:
  1. Sensor Subsystem Responsible for acquiring physiological and motion data.
    - The MAX30102 sensor measures heart rate and blood oxygen saturation.
    - The MPU6050 accelerometer and gyroscope detect head movements and posture changes indicating fatigue.
  2. Processing Subsystem Powered by the ESP32 microcontroller, which:
    - Reads and processes sensor data.
    - Performs feature extraction and AI inference using a pre-trained TinyML model deployed through Edge Impulse.
    - Manages data logging and system control logic.
  3. Feedback and Alert Subsystem Provides real- time
  user feedback using:
  - Piezo buzzer for auditory alerts.
  - OLED display (SSD1306) for on-screen updates
    
    of heart rate, SpO, and fatigue level.
  - RGB LED indicators for visual alerts (Green
  Normal, Yellow Warning, Red Critical).
2. Data Flow Architecture
  
  The data flow within PulseDrive follows a structured sequence that enables low-latency real-time monitoring and response:
  
  Sensors ESP32 Data Buffer TinyML Model
  
  Classification Alerts / OLED Display
  
  Optional Cloud Sync
  - Sensors: Capture continuous physiological and motion data.
  - Data Buffer: Aggregates readings for stable AI inference.
  - TinyML Model: Classifies the drivers condition as
    
    Alert or Drowsy.
  - Alert Subsystem: Issues warnings through LED, buzzer, and on-screen display.
  - Cloud Sync (Optional): Enables centralized fleet- level monitoring for predictive analytics.
    
    This streamlined data flow ensures efficient real- time detection, low computational overhead, and seamless integration with fleet management platforms.
Workflow

The system was developed and tested following a structured, iterative workflow to ensure functionality, reliability, and real-world applicability.
1. Hardware Prototyping
  - The prototype was built on a breadboard with all components interconnected as per the designed circuit diagram.
  - Sensors (MAX30102 and MPU6050) communicate with the ESP32 via the I²C protocol.
  - The OLED display, buzzer, and LED modules are connected to dedicated GPIO pins for output functions.
  - Power is supplied through a regulated 5V/3.3V supply to ensure stable operation during testing.
2. Software Implementation
  
  The entire system logic is implemented using Arduino IDE, integrating multiple libraries for sensor communication, AI inference, and data visualization.
  
  The master Arduino sketch performs the following major tasks:
  1. Sensor initialization and continuous data acquisition.
  2. Data buffering and preprocessing for model compatibility.
  3. Running AI inference using the embedded TinyML classifier (run_classifier() function).
  4. Activating alerts based on classification results.
  5. Updating OLED display with current readings and driver state.
  6. Logging sensor and classification data to SD card for later analysis.
  This unified firmware design ensures optimized execution, reduced latency, and high responsiveness to real-time physiological changes.
3. Testing and Validation
  
  The system underwent iterative testing across various simulated driving scenarios to evaluate:
  - Sensor Accuracy: Cross-verified with medical- grade pulse oximeters.
  - Model Performance: Validated classification accuracy and false detection rates.
  - Response Time: Measured delay between fatigue detection and alert activation.
  - Power Efficiency: Assessed for continuous operation under vehicle power conditions.
    
    The prototype successfully detected driver fatigue states with high reliability, providing instant alerts without noticeable delay. These results demonstrate the feasibility of PulseDrive for real-time deployment in buses and heavy vehicles.
4. Adaptability and Scalability
  
  The modular architecture allows easy adaptation of PulseDrive for diverse applications:
  - Fleet Monitoring: Integration with IoT-based cloud dashboards for centralized driver health analysis.
  - Vehicle Integration: Seamless embedding within dashboards or steering systems.
  - Custom Analytics: Expansion with additional sensors (ECG, temperature, etc.) for comprehensive monitoring.

This scalability makes PulseDrive suitable for future smart transportation ecosystems and AI-assisted mobility frameworks.

VI ARCHITECTURE DIAGRAM

An architecture diagram shows the high-level "blueprint" of The project. It explains what the main components are and, most importantly, how data and commands flow between them.

Sensor Layer

Collects real-time physiological and motion data using:
- MAX30102: Measures heart rate, HRV, and
  
  SpO via PPG.
- MPU6050: Detects motion and fatigue indicators using accelerometer and gyroscope data. Data is transmitted through I²C to the ESP32 for processing.
Processing Layer

ESP32 serves as the core processor handling:
- Data acquisition from sensors.
- TinyML inference for fatigue classification using fused data frm both sensors. Output labels: Alert or Drowsy.
Feedback Layer

Executes alerts and displays information:
- Buzzer: Immediate auditory warning.
- OLED Display: Shows HR, SpO, and system status.
- RGB LED: Indicates alert level (Green Normal, RedCritical).
Cloud Layer

Optional remote monitoring using Wi-Fi or Bluetooth.
- Sends data to cloud via MQTT/Firebase/Blynk.
- Enables mobile or fleet dashboards for real-time alerts and analytics.

VII.MODEL EVALUATION AND TESTING

Machine Learning Model Evaluation

The TinyML classification model was evaluated in Edge Impulse Studio using a 20% unseen test dataset. The models goal was to accurately classify driver states as Alert or Drowsy.
1. Confusion Matrix Overview
  - True Positive (TP): Correctly detected
    
    Drowsy.
  - True Negative (TN): Correctly identified
    
    Alert.
  - False Positive (FP): Incorrectly signaled fatigue (false alarm).
  - False Negative (FN): Missed detecting fatigue (critical error).
2. Performance Metrics
  - Accuracy 93.5%: Model predictions correct most of the time.
  - Precision 94.8%: Alerts were reliable and minimized false alarms.
  - Recall 92.0%: Successfully detected most drowsiness events.
  - F1-Score 93.4%: Balanced precision and recall, showing model stability.
    
    These results confirm the TinyML model is both accurate and computationally efficient for real- time driver monitoring.
Real-World Testing (Simulation)

To validate real-world performance, a

simulated driving test was conducted:
- Setup: Prototype mounted in a stationary vehicle; subjects wore sensors.
- Protocol:
  - 3 mins Alert Phase: Normal posture and movements.
  - 2 mins Drowsy Phase: Simulated head- nods and posture slumps.
- Participants: Three users performed 5- minute sessions each.
Results and Observations

AI Accuracy: 93.5% (lab) 8890% (real- world) due to sensor noise.
Latency: <1.5 seconds (meets real-time alert goal).
False Positives: Minimal; system remains user- friendly.
False Negatives (~11%): Highlight need for more diverse training data.

VIII RESULT DISCUSSION

The evaluation results show that PulseDrive AI- Powered Driver Health and Fatigue Monitoring System performs

effectively with 93% accuracy, low latency (<1.5s), and

minimal false alarms under simulated conditions.

The main limitation is that testing was simulated, not conducted on genuinely fatigued drivers or in moving vehicles, where real-world factors like vibrations and lighting may impact accuracy.

Overall, the system proves technically sound and practical for real-time fatigue detection. Future work will include real- world testing, model refinement, and cloud integration to improve robustness and adaptability.

CONCLUSION AND FUTURE PROSPECTS
Future improvements include:
1. Real-world testing with genuine fatigue data for better model accuracy.
2. Miniaturized PCB design for practical, in-vehicle use.
3. Noise filtering (e.g., Kalman filters) to handle vibration artifacts.
4. Cloud and GSM integration for remote monitoring and emergency alerts.
5. Expanded AI models to detect additional states like stress or distraction.
In conclusion, PulseDrive represents a scalable foundation for smart, AI-driven mobility safety solutions that enhance driver well-being and reduce road accidents.
RESERCHES
References

Espressif Systems, ESP32-WROOM-32 Datasheet, Rev. 3.3, 2022.
[Online]. Available:

https://www.espressif.com/sites/default/files/documentation/esp32- wroom-32_datasheet_en.pdf
Analog Devices, MAX30102: High-Sensitivity Pulse Oximeter and Heart-Rate Sensor, Rev. 2, 2019. [Online]. Available: https://www.analog.com/media/en/technical-documentation/data- sheets/MAX30102.pdf
TDK InvenSense, MPU-6000 and MPU-6050 Register Map and

Descriptions, Rev. 4.2, 2013. [Online]. Available:

https://invensense.tdk.com/wp-content/uploads/2015/02/MPU- 6000-

Register-Map1.pdf
S. M. R. Islam, M. A. Kauser, A. Al Mueed, and M. S. R. Swarna, Drowsiness Detection System using ESP32 and MPU6050, Proc. Int. Conf. on Advancement in Electrical and Electronic Engineering (ICAEEE), 2022, pp. 14.
P. Warden and D. Situnayake, TinyML: Machine Learning with TensorFlow Lite on Arduino and Ultra-Low-Power Microcontrollers, Sebastopol, CA: OReilly Media, 2019.
J. M. D. J. Reddi, Edge Impulse: An MLOps Platform for Tiny Machine Learning, Proc. MLSys Conf., 2022.
A. K. Sahani and A. K. Singh, Driver Drowsiness Detection Based on Physiological Signals: A Review, Proc. Int. Conf. on Computational Intelligence and Computing Applications (ICCICA), 2021, pp. 16.
L. Li, K. Lin, and X. Wang, A Novel Driver Drowsiness Detection System Based on Heart Rate Variability and Head Posture, Proc. IEEE Int. Conf. on Vehicular Electronics and Safety (ICVES), 2019,

pp. 16.
U. A. Contardi, M. Morikawa, B. Brunelli, and D. V. Thomaz, MAX30102 Photometric Biosensor Coupled to ESP32-Webserver Capabilities for Continuous Point-of-Care Oxygen Saturation and Heart Rate Monitoring, Engineering Proceedings, vol. 16, no. 1, p. 9, 2022.
S. H. M. Al-Shayea, Driver Drowsiness Detection System Using IoT, Int. J. of Advanced Computer Science and Applications, vol. 11, no. 9, 2020.
M. Shahverdy, M. Fathy, R. Berangi, and M. Sabokrou, Driver Behavior Detection and Classification Using Deep Convolutional Neural Networks, Expert Systems with Applications, vol. 149, Jul. 2020, Art. no. 113240