ARCHIVES

The increasing frequency of fire-related accidents in both urban and natural environments highlights the urgent need for rapid and reliable detection systems. Traditional fire detection approaches, primarily based on heat sensors and human intervention, often result in delayed responses and limited accuracy. This project introduces a deep learning-based flame and smoke detection framework using the YOLOv5s architecture integrated with ODConvBS (Omni-Dimensional Convolution with Batch Selection). The proposed system leverages computer vision to detect early signs of fire and smoke directly from images and video feeds, ensuring faster response times. A carefully curated dataset of flame and smoke images was collected, annotated, and augmented to enhance model robustness across diverse environments. Performance validation using precision, recall, and F1-score metrics demonstrates a reduction in false positives compared to conventional methods. A Streamlit-based interface was developed, enabling real-time monitoring, analysis, and deployment across surveillance systems like CCTV networks or smart city platforms. The system requires modest computational resources, making it scalable and practical for real-world applications in buildings, factories, and forests. The outcomes highlight the system's ability to deliver accurate, efficient, and real-time detection, paving the way for improved safety, disaster prevention, and expanded applications in hazard monitoring using AI.