Deep Neural Network Based Anomaly Detection For Smart Surveillance System

Abstract

Searching for anomalies in surveillance film is a practise that is becoming more and more popular. Anomaly detection is the procedure of spotting irregularities in data. When we talk about “abnormal data,” we mean data that drastically differs from its typical behaviour. Fraud detection, malfunction detection, and intrusion detection are just a few of the uses for anomaly detection. You Only Look Once is a method for quickly recognising objects (YOLO). It is an object identification system that is capable of quickly locating objects in images, real-time coverage, and video streams. An instant object recognition system is the YOLO Convolutional Neural Network (CNN). CNNs are classifier-based systems that can identify patterns in images and interpreting them as collections of well-organized data. Both the intersection over union (IOU) and mean average precision (mAP) values of YOLOv3 are quick and exact. This detection method is significantly slower than other detection methods that get comparable results. This study suggests utilising a modified YOLOv3 algorithm to track things automatically and notify people if anything seems off. The performance of the algorithm was also contrasted with that of other algorithms, including CNN and decision trees. Our system continuously records live video while abnormal things like fire and weapons are identified. High detection precision and quick processing are features of our technology.