EMERGENCY DETECTION USING COMPUTER VISION TECHNOLOGY: AN AI-BASED FRAMEWORK FOR REAL-TIME SAFETY MONITORING

Abstract

The rise in accidents, emergencies, fires, and safety issues demands smart automation systems that can help detect and respond to emergencies. Conventional surveillance systems depend on manual observation, which is inefficient, prone to human errors, and incapable of providing instant alarms. This paper suggests an AI-enabled framework for detecting emergencies using computer vision for instant safety monitoring. The proposed framework has four parts: video acquisition, preprocessing and object detection, classification of emergencies, and alerts generation. A huge dataset of 25,000 images and frames containing fire, smoke, accidents, falls, and violence in crowds was collected from publicly available data resources including Kaggle, UCSD Anomaly Detection, and live CCTV cameras. Data was preprocessed using frame extraction, resizing, denoising, and augmentation techniques. Deep learning algorithms such as YOLOv8, Convolutional Neural Network (CNN), and Long Short-Term Memory (LSTM) were designed using Python and OpenCV software. The performance of the model was assessed on the basis of accuracy, precision, recall, F1-score, and response time. It can be seen from the experimental findings that the YOLOv8 model has proved superior to other models in achieving an accuracy of 95.4%, a precision of 96.1%, a recall of 94.7%, and an average detection time of 1.1 seconds, which is much higher compared to the traditional manual surveillance system. This intelligent system not only ensures public safety but also scalability in smart cities and flexibility in diverse emergencies. Contributions to Computer Science, AI & Data Science, and Information Technology through this study have been made in the form of developing this intelligent surveillance system for emergency management.