Suspicious Activity Detection using LSTM and MobileNetV2

Abstract

For the prevention of security issues in publicly accessible areas, it is necessary to integrate computer vision and AI into an automatic video identification system. In detecting abnormal behaviour, traditional surveillance methods are insufficient and a system needs to be automated. The objective of the project is to use deep learning techniques, especially CNN models, for analysing video footage posted on a web site in order to make surveillance more efficient. In addition, it involves segmenting the video into frames, extracting features using MobileNetV2 and identifying irregular or suspicious activities. The system's functions include background and foreground extraction and anomaly detection that allows for a distinct distinction in the behaviour of normal and irregular activities on surveillance video. The study seeks to bridge the gap between surveillance technology by involving computer vision, image processing and artificial intelligence so as to be able to quickly identify unusual actions on video. In addition, when detecting potential security threats, it ensures that timely alerts are sent via email. This research demonstrates the importance of addressing emerging security challenges in today's cities, contributing to enhancing surveillance systems.