A Short Review of Deep Learning Methods for Understanding Group and Crowd Activities

The human behavior analysis in groups and crowds is an important research area in different fields of science (e.g. sociology, psychology, computer science). Specifically, the automated understanding of the activities of both, groups and crowds, remains a very challenging problem from sensors to artificial intelligence techniques. Currently, deep learning (DL) methods are achieving great results that are revolutionizing the way to deal with artificial vision problems. The objective of this paper is to briefly review the state of the art in this area of machine learning to solve problems related to image processing and automatic understanding of group and crowd activities detected in video sequences and images. First of all, the review presents the general characteristics of the three main neural networks and their variants: Convolutional Neural Networks (CNNs), Autoencoders (AEs) and Recurrent Neural Network (RNNs). Later, the paper highlights the research work in DL methods for understanding separately group and crowd in sequence of images and the main datasets used for this task. Finally, the different levels of understanding and the number of people considered in the group and crowd is discussed proposing a classification of the analyzed methods using both variables and identifying open challenges in the area.