Understanding Neural Networks in Surveillance Systems

Abstract The AI‑driven neural networks of surveillance systems have the capability of making automatic decisions and identifying patterns. This chapter goes further about artificial neural networks (ANN), deep neural networks (DNN) and their variations, convolutional neural networks (CNN), and long short‑term memory (LSTM) networks. It discusses how video is presented in these systems to their feeds, how anomalies are detected within it, and how they assist in improving video security monitoring. In addition, the chapter discusses how a model in the surveillance application can be made more accurate using neural network training procedures, optimization techniques, and transfer learning techniques. It is one of the measures of the cost of processing and the bias in the training data and adversarial attacks on the AI model. The third part of this thesis describes, thirdly, the applications of neural networks in smart surveillance, facial recognition, and crowd behavior analysis.