Web-Based Road Accident Detection System Using YOLOv11 Deep Learning Model and Flask Framework

Muhammad Imron Maulana¹, Arjuni Budi Pantjawati¹

1. ¹ Indonesia University of Education

International Symposium on Materials and Electrical Engineering (ISMEE), 2025

This research aims to develop a traffic accident and armed criminal activity detection system using the YOLOv11 deep learning model integrated with a Flask-based web platform. The system is designed to provide real-time notifications to improve emergency response unit responsiveness. The research methodology employs Research and Development (R&D) with a Waterfall approach that includes hybrid dataset collection, preprocessing, YOLOv11n model training, as well as web system development and integration. The dataset consists of 2,301 augmented images divided into 85% training, 10% validation, and 5% testing. Model evaluation results demonstrate high performance with precision 97.1%, recall 98.2%, mAP50 98.9%, and mAP50-95 87.9%. In web-based system testing, local video achieved 91.11% accuracy, average detection speed of 61.14 ms, frame rate of 8.66 FPS, and latency of 63.4 ms. For real-time IP CCTV, accuracy reached 64.44%, detection speed of 65.72 ms, frame rate of 6.57 FPS, and total latency of 119.1 ms. The alert system successfully triggered audio notifications and saved incident evidence with 100% accuracy. The research conclusion affirms the system's effectiveness in rapidly detecting emergency incidents, although it requires more diverse dataset development and computational infrastructure optimization to improve real-time performance.

DOI10.1109/ismee68179.2025.11473163Copy