ACCITRACK: Real-Time Accident Detection and Emergency Response System using YOLOv8

ACCITRACK: Real-Time Accident Detection and Emergency Response System using YOLOv8

Parvathy S Pillai (1), Adithya Raj (1), Shuhood K M (1), Binni N U (1), Dr. Deepa S Kumar (2), Prof. Jisha A K (3)

(1)Department of CSE, College of Engineering Munnar, Kerala, India

(2)Associate Professor, Department of CSE, College of Engineering Munnar, Kerala, India

(3)Assistant Professor, Department of CSE, College of Engineering Munnar, Kerala, India

Abstract – Delayed emergency response during the golden hour following road trafc accidents leads to signicant loss of life. ACCITRACK is an AI-based real-time accident detection and response system that utilizes live video feeds and integrates the YOLOv8 deep learning model for automatic detection of road accidents and assessment of their severity. Upon detection, the system generates instantaneous alerts along with location details and for-wards them to nearby hospitals and police stations through a Python-based communication mechanism. The system incorporates an ESP32 microcontroller integrated with a GPS module to obtain real-time location information. By enabling automated detection and rapid alert transmis-sion, the proposed system minimizes response time and reduces reliance on manual reporting. Overall, the system enhances emergency response efciency and contributes to improved road safety, making it suitable for smart trafc management and smart city applications.

Index TermsAccident Detection, YOLOv8, Computer Vision, ESP32, GPS, Emergency Alert System, Smart Cities

1. INTRODUCTION

   Road trafc accidents are a major global safety concern, causing millions of injuries and fatalities each year. A signicant number of deaths occur due to delays in providing medical assistance during the critical golden hour. Traditional accident reporting methods rely on manual intervention, such as eyewitness reports or emer-gency calls, which may be delayed or unavailable, espe-cially in remote areas. With advancements in Articial Intelligence and Computer Vision, automated accident detection has emerged as an effective solution. Deep learning models, particularly the YOLO family, enable real-time detection of accidents from video streams with high accuracy.

   The proposed system is an AI-based real-time acci-dent detection and emergency response system that uses

   YOLOv8 to identify accidents from live or recorded video feeds. It classies accidents into moderate and se-vere categories to prioritize emergency response, where moderate cases notify police, and severe cases alert both police and hospitals. The system integrates an ESP32 microcontroller with a GPS module to obtain real-time location data and uses a Python-based communication module to send alerts. A web dashboard is also included for monitoring and data visualization. By combining detection, location tracking, and automated alerts, the system reduces response time and enhances road safety.

2. LITERATURE SURVEY

   Recent advancements in deep learning have enabled efcient accident detection using computer vision tech-niques [1], [2]. Traditional methods based on sensors and manual reporting were less reliable and time-consuming [3]. Modern approaches use YOLO-based models for real-time detection due to their high accuracy and speed [15].

   Many systems also integrate GPS and communication technologies to provide location-based alerts to emer-gency services [4], [10]. Some studies further include severity classication to improve response prioritization [8].

   However, existing systems often lack full integration of detection, classication, location tracking, and alert mechanisms [9], [13]. The proposed ACCITRACK sys-tem addresses these limitations by combining all these features into a unied framework for faster and more reliable emergency response.

3. PROBLEM STATEMENT

   To develop a real-time AI-driven accident detection system using YOLO-based computer vision and GPS technology that can automatically detect road accidents,

   analyze their severity, and send immediate alerts to nearby authorities to enable rapid emergency response.

4. OBJECTIVES

   • To develop an AI-based accident detection system using YOLOv8 for real-time identication.

   • To classify accidents into moderate and severe cate-gories.

   • To integrate GPS for accurate location tracking.

   • To design an automated alert system for notifying hospitals and police.

   • To develop a web dashboard for monitoring and visu-alization.

5. EXISTING SYSTEM

   Traditional accident detection systems rely on manual reporting methods such as eyewitness communication or emergency calls, which often lead to delays in response time [3]. These approaches are not always reliable, especially in remote areas. Sensor-based systems using accelerometers and vibration sensors have also been developed [7]. Recent approaches utilize deep learning models such as YOLO for accident detection from video streams [12], [15]. However, most systems lack full integration and severity classication.

6. PROPOSED SYSTEM

   1. System Overview

      The proposed system, ACCITRACK, is an AI-based real-time accident detection and emergency response system designed to reduce response time and improve road safety. The system processes live or recorded video input and uses the YOLOv8 deep learning model to detect accident occurrences.

      Once an accident is detected, the system classies it into moderate or severe categories. Based on the severity, alerts are generated accordingly. Moderate accidents trig-ger notications to nearby police stations, while severe accidents send alerts to both police and hospitals for immediate action.

      The system integrates an ESP32 microcontroller with a GPS module to obtain real-time location data. A Python-based communication module is used to transmit alerts through SMS or API services. Additionally, a web dashboard is provided for monitoring accident data and maintaining records.

   2. System Architecture

   The system consists of multiple modules where video input is processed using OpenCV and passed to the

   YOLOv8 model for accident detection. A classication module categorizes the accident, and location data is obtained using ESP32 with GPS. Alerts are then sent to the appropriate authorities, and a web dashboard is used for data visualization and storage.

   Fig. 1. System Architecture of ACCITRACK

7. METHODOLOGY

   1. Dataset Preparation

      The dataset consists of accident and non-accident images collected from public sources, including the Roboow dataset [17].. It includes different categories such as moderate and severe accidents. The dataset is prepared to train the model for accurate accident detection.

   2. Data Preprocessing

      Preprocessing techniques such as image resizing, nor-malization, and annotation using bounding boxes are applied to improve model performance. The dataset is also split into training and testing sets.

   3. Model Development

      he YOLOv8 model is used for accident detection due to its high accuracy and real-time processing capability. Transfer learning is applied using pre-trained weights,

      and the model is traind on the prepared dataset. Per-formance metrics such as precision, recall, and loss are monitored during training.

   4. System Workow

      The system takes video input and processes it frame by frame. The trained YOLOv8 model detects accidents and classies them into moderate and severe categories. Based on the classication, location tracking and alert generation processes are triggered.

      Fig. 1. System Workow

   5. Real-Time Accident Detection

      The trained model processes video streams in real time to detect accident events. Once detected, the system clas-sies the accident based on severity to enable prioritized emergency response.

   6. Alert Generation and Location Tracking

      The system integrates an ESP32 microcontroller with a GPS module to obtain real-time location data. A Python-based communication module sends alerts based on severity. Moderate accidents notify police, while severe accidents notify both police and hospitals.

   7. Web Dashboard

   A web-based dashboard is separately implemented for both hospital and police to view accident de-tails such as location, severity, and time. It also

   stores historical data for monitoring and analysis.

   Fig. 2. Police Dashboard of ACCITRACK

   Fig. 3. Hospital Dashboard of ACCITRACK

8. RESULTS AND PERFORMANCE ANALYSIS

   The proposed ACCITRACK system was evaluated using a dataset consisting of accident and non-accident images. The YOLOv8 model was trained and tested to assess its ability to detect accidents in real-time video streams.

   The performance of the model was evaluated using stan-dard metrics such as accuracy, precision, recall, and F1-score. The system achieved high accuracy in detecting accident events and demonstrated reliable performance in differentiating between moderate and severe cases.

   1. Performance Metrics

      • Accuracy: 97%

      • Precision: 98.5%

      • Recall: 98.4%

      • F1-Score: 98.4%

        The results indicate that the model is effective in iden-tifying accident scenarios with minimal false detections. The high precision ensures fewer false alarms, while good recall ensures most accident cases are detected.

   2. Detection Performance

      The system successfully detects accidents in real-time video streams with minimal delay. It accurately identies

      accident regions and classies them based on severity. The integration of YOLOv8 enables fast processing and high detection speed, making the system suitable for real-time applications.

      Fig. 2. Performance Metrics of the Proposed System

   3. System Evaluation

   The system was tested under different conditions, in-cluding varying lighting and trafc scenarios. The re-sults show that the system performs consistently well, although slight variations may occur in complex envi-ronments.

   The integration of GPS and alert mechanisms ensures that emergency notications are sent promptly after de-tection. The web dashboard effectively displays accident data and supports monitoring and analysis.

   Overall, the proposed system demonstrates efcient per-formance in real-time accident detection, classication, and alert generation, making it suitable for practical deployment.

9. DISCUSSION

   The system provides an effective approach for real-time accident detection using the YOLOv8 model. The system is capable of identifying accident scenarios from video streams with good accuracy and processing speed. The inclusion of severity classication into moderate and severe categories supports better prioritization in emergency response.

   The integration of GPS-based location tracking and au-tomated alert generation enhances the practical usability of the system. The web dashboard offers a convenient interface for monitoring accident data and maintain-ing records, contributing to improved situational aware-ness.Certain limitations are observed under challenging conditions such as low-light environments, poor video

   quality, and dense trafc scenarios, which may affect detection performance. The system also depends on sta-ble network connectivity for real-time alert transmission and dashboard updates. Additionally, hardware integra-tion may introduce constraints during real-world deploy-ment.Overall, the system provides a reliable framework for accident detection and emergency response. With further improvements, it can be extended for large-scale deployment in intelligent transportation systems and smart city applications.

10. CONCLUSION

    The ACCITRACK system presents an effective solution for real-time accident detection and emergency response using deep learning and IoT technologies. The inte-gration of the YOLOv8 model enables accurate and fast detection of accident events from video streams. The classication of accidents into moderate and severe categories supports prioritized emergency response.The incorporation of an ESP32 microcontroller with a GPS module ensures accurate location tracking, while the alert system enables timely notication to police and hos-pitals. The web dashboard further enhances the system by providing real-time monitoring and data visualization.

    The results demonstrate that the system achieves high accuracy and reliable performance in detecting accidents and generating alerts. By reducing response time and minimizing reliance on manual reporting, the system contributes to improved road safety and supports the development of intelligent transportation systems.

11. FUTURE WORK

The proposed system can be further enhanced by in-tegrating advanced deep learning models to improve detection accuracy under challenging conditions such as low light, weather disturbances, and dense trafc. The use of larger and more diverse datasets can also help in improving model generalization.

Future improvements may include full hardware imple-mentation with ESP32, GPS, and GSM modules for in-dependent operation without relying on external systems. Integration with cloud platforms can enable real-time data storage, analytics, and large-scale deployment.

Additionally, incorporating multi-sensor data such as IoT sensors, LiDAR, or radar can improve detection reliabil-ity. The system can also be extended to include automatic emergency call services and direct integration with smart city infrastructure for enhanced trafc management and faster emergency response.

REFERENCES

1. R. Sabitha, D. Tejashree, and G. Yaashish, Real-Time Accident Detection and Response System Using Resnet Algorithm, in 2024 IEEE 16th International Conference on Computational Intelligence and Communication Networks (CICN), 2024, Art. no. 10847502. doi: 10.1109/CICN63059.2024.10847502.

2. N. Annela and T. N. Nishith, Real-Time Accident Detection Leveraging Deep Learning for Enhanced Road Safety, Int. J. Eng. Res. Technol. (IJERT), vol. 13, no. 12, Dec. 2024.

3. P. Mahendhar, R. Nihal, and P. Vikas, Articial Intelligence for Accident Detection and Response, J. Emerg. Technol. Innov. Res. (JETIR), vol. 6, no. 3, pp. 711, Mar. 2019.

4. T. S. L. Pasad, V. Lokesh, B. Sampath, and B. Hariteja, Acci-dent Detection and Alert System, Global J. Eng. Innovations & Interdiscipl. Res., Jun. 2025.

5. H. Ghahremannezhad, H. Shi, and C. Liu, Real-Time Acci-dent Detection in Trafc Surveillance Using Deep Learning, Unpublished Conf. Paper, Jun. 2022.

6. H. Ghahremannezhad, H. Shi, and C. Liu, Real-Time Acci-dent Detction in Trafc Surveillance Using Deep Learning, Unpublished Conf. Paper, Jun. 2022.

7. S. P, K. B, and K. M, SMART TRAFFIC ACCIDENT DE-TECTION AND AUTOMATED EMERGENCY RESPONSE SYSTEM USING OBJECT DETECTION ALGORITHM, Int.

   J. Res. Trends Innovation (IJRTI), vol. 10, no. 4, pp. a522a527, Apr. 2025.

8. M. G. Shaikh and K. D. P, Real-Time Accident Detection and Response System for Enhanced Road Safety using Deep Learning, Int. J. Res. Pub. Reviews, vol. 6, no. 3, pp. 62806288, Mar. 2025.

9. B. R. Duganavar, S. A. Vardhan, C. Chinchanikar, N. Magadum, and H. H. K. Harish, Smart Accident Detection and Emergency Response System for Enhanced Road Safety, Int. J. Innovative Sci. Res. Technol. (IJISRT), vol. 9, no. 5, May 2024. doi: 10.38124/ijisrt/IJISRT24MAY394.

10. H. R. Prakash, V. M., R. Seetharaman, and V. Balaji, Real-Time Accident Detection and Notication System, Int. J. Innovative Res. Sci. Eng. Technol. (IJIRSET), vol. 13, no. 5, May 2024. doi: 10.15680/IJIRSET.2024.1305116.

11. H. M. S. Bandara et al., Real Time Accident Detection and Emergency Response Using Drones, Machine Learning and LoRa Communication, Int. J. Adv. Comput. Sci. Appl. (IJACSA), vol. 16, no. 6, 2025.

12. P. S. Raju et al., Accident Detection System on Roads using Convolutional Neural Network, Int. J. Adv. Res. Sci. Commun. Technol. (IJARSCT), vol. 4, no. 4, pp. 415420, Apr. 2024. doi: 10.48175/IJARSCT-17471.

13. M. Nemade, Y. Shah, P. Shah, and R. Patil, AcciSenseReal-time Accident Detection and Emergency Response System, Int.

    J. Advance Res., Ideas and Innovations in Tech., vol. 9, no. 6,

    pp. 15, 2023.

14. M. S. Aren, M. I. S. Mahin, and F. A. Mily, Real-time rapid accident detection for optimizing road safety in Bangladesh, Heliyon, vol. 11, 2025, Art. no. e42432. doi: 10.1016/j.heliyon.2025.e42432.

15. A. Darokar, A. Talware, A. Jadhav, D. Guntiwar, S. Dandhare, and D. Dandekar, Accident Detection System Using YOLOv8 and CNN Algorithms, Int. Res. J. Modernization Eng., Tech. Sci., vol. 05, no. 12, pp. 35683572, Dec. 2023.

16. F. Liang, Deep learning based trafc accident detection, Sam Houston State Univ., Huntsville, TX, USA, Tech. Rep. IHS/CR-2024-1041, 2024. doi: 10.17605/OSF.IO/Q6BKH.

17. Roboow, Accident Detection Dataset, [Online]. Available: https://universe.roboow.com/. Accessed: 2026.