Аннотация
In recent years there was significant progress in analyzing video recordings to identify criminal and suspicious behavior. Major types of machine learning architectures used in these studies are 2D and 3D Convolutional Neural Network (CNN) architecture. This article aims to review a notion of general CNN’s as well as their various modifications used for crime detection from video footage. In addition to the introduction to architectural details, this paper is focused on studying and comparing methods used in various studies on detecting criminal behavior using CNN neural architecture. Finally, the paper presents possible directions for future research in the given domain.