A Systematic Review on Evaluation of Surveillance System for Unusual Behavior Using Artificial Intelligence

Abstract

The latest advancements in Artificial Intelligence and Computer Vision have significantly enhanced video surveillance capabilities. The primary purpose of a Video Surveillance System (VSS) is to improve security and privacy through the meaningful analysis of video frames. VSS involves the supervision or remote monitoring of specific events using cameras and recorders. Although many current systems rely on conventional image processing and some incorporate machine learning, challenges such as dynamic scenes, performance improvements, and weather conditions impact accuracy and efficiency. This work proposes an efficient approach designed to be both fast and accurate, addressing these challenges. Intelligence will be integrated into the system to filter and interpret data captured by cameras, where human observers may struggle to assess developments in real-time. The proposed method will analyse video to detect and recognize humans, vehicles, attributes, and abnormal events. Combining AI, Computer Vision, and the Internet of Things (IoT) can lead to numerous benefits, including reduced human fatigue, enhanced audio and video analysis, movement pattern recognition, gesture tracking, and behavioural analysis, ultimately conserving human resources. Our aim is to develop a system capable of real-time object detection and classification using a live camera feed. The primary objective is to accurately identify and classify objects into two main categories: humans and vehicles. The system includes following approaches which was used to determine the objective of the project: Feature Extraction, Anomaly Detection Model, Training and Validation, Integration and Alert Mechanism. The obtained results will be compared using different deep learning approaches to ensure optimal performance.