Year: 2025
Type: article
Abstract: - The growing complexity of urban infrastructure and the frequency of accidents and fire hazards highlight the limitations of conventional CCTV systems, which rely heavily on manual monitoring and delayed responses. This study presents a real-time, AI-powered surveillance system capable of detecting vehicular crashes, fire, and smoke from live CCTV footage. At its core, the system employs the YOLOv8 object detection model, leveraging its speed and accuracy for high-performance visual analytics. Incoming video frames are processed through a deep learning pipeline, with event data transmitted via a Flask-SocketIO backend to a web-based frontend. MongoDB is used for structured alert logging, while the dashboard provides live annotated streams and real-time notifications. Evaluation results show crash detection precision exceeding 90%, alert latency under one second, and reliable performance across diverse scenarios. The system also supports multi-camera feeds and SMS/email alerting through external APIs. By integrating deep learning, computer vision, and web technologies, this solution significantly enhances emergency responsiveness and scalability in surveillance infrastructure. Key Words: YOLOv8, Computer Vision, Real-Time Detection, Fire Monitoring, Crash Surveillance.
DOI: 10.55041/ijsrem48718
Author: Himanshu Jaiswara
Language: English
Subfield: Safety, Risk, Reliability and Quality
Field: Engineering
Domain: Physical Sciences
Open Access status: bronze
OpenAlex ID: W4410744978