Paper Title
AI-Driven Real-Time Surveillance System for Automated Abnormal Behaviour Detection and SOS Alert

Abstract
The surveillance systems used nowadays require human intervention for monitoring, which is inefficient because errors may occur. The paper proposed shows the solution for a real-time surveillance system that is done using AI and ML, which is used to identify activities like road accidents, fire accident and violence that need immediate attention. To ensure the proposed system, we use object detection and behaviour analysis from the video that is fed. To provide a better performance, the proposed system uses the YOLOv8 algorithm for object detection and the CNN-LSTM model for behavioural analysis. The combo used here provides efficient extraction of behavioural patterns from the frames that are received from the video clips. The video clips that are received as a dataset are extracted into frames from analysis of various behaviours that are categorised into normal and abnormal. If any abnormal behaviour is found, an SOS alert notification is automatically generated from the proposed system, which includes the location and a picture of the incident from the video clip. The SOS message is mainly focused on the rapid response needed for an emergency situation. This alert SOS is integrated in the proposed system using the Twilio API with WhatsApp Sandbox configuration for better communication. The proposed model is also evaluated with various performance metrics like F1 score, accuracy and confusion matrix. Keywords - YOLOv8 Object Detection, CNN-LSTM Model, Abnormal Behaviour, Automated Surveillance System.