International Journal of Contemporary Research In Multidisciplinary, 2023;2(6):16-23
A Comprehensive Literature Survey for Crowd Scene Analysis techniques
Author Name: Faisel G. Mohammed, Abbas F. Nori, Noor N. Thamer
Paper Type: research paper
Article Information
Abstract:
Studying how people behave in crowded places is crucial for controlling the spread of diseases like COVID-19. This survey article presents an in-depth examination of crowd scene analysis methods, including crowd counting and crowd activity detection. It covers the spectrum up to contemporary deep learning techniques, which are often overlooked in current studies that focus mainly on traditional approaches or specific aspects.
The article introduces the innovative concept of Crowd Divergence (CD) evaluation, which is a matrix for evaluating crowd scene analysis methods based on information theory. CD quantifies the agreement between expected and observed crowd count distributions, unlike traditional measurements. The paper makes three significant contributions: examining available crowd scene datasets, using CD for a thorough evaluation of techniques, and providing a comprehensive review of crowd scene methodologies.
The investigation begins with conventional computer vision methods, including density estimates, detection, and regression strategies. As deep learning advances, convolutional neural networks (CNNs) become effective tools, as demonstrated by new models like ADCrowdNet and PDANet, which use attention mechanisms and structured feature representation. A variety of benchmark datasets, such as ShanghaiTech, UCF CC 50, and UCSD, are analyzed to evaluate algorithmic effectiveness.
Crowd scene analysis is a fascinating and challenging topic in computer vision, with numerous applications ranging from crowd control to security surveillance. This survey article offers a comprehensive overview of crowd scene analysis, bringing together multiple approaches under a single heading and presenting the CD measure to ensure reliable evaluation. It provides a complete resource for researchers and practitioners through a detailed investigation of methods, datasets, and cutting-edge evaluation techniques, paving the way for improved crowd scene analysis techniques in various fields.
Keywords:
Crowd behavior analysis, Crowd scene methodologies, Crowd Divergence (CD) evaluation, Deep learning techniques, Benchmark datasets, Crowd control and security
Introduction:
The study of crowd scene analysis involves examining the behavior of people in groups in the same physical space [1]. It usually entails counting the number of people, in regions tracking their movements and identifying their behaviors. This type of analysis has applications. One such application is controlling the spread of COVID 19 by guaranteeing separation in public spaces such as malls and parks [2]. Additionally, it helps to maintain security on Muslim pilgrimages, carnivals, New Year's Eve celebrations, and sporting events. [3, 4]. Surveillance camera systems may detect behaviors within groups of individuals using automatic crowd scene analysis [5]. Additionally analyzing crowd scenes in places like train stations, supermarkets and shopping malls can provide insights, into crowd movement patterns. Identify design flaws. These studies contribute to improving safety considerations [6, 7]. As was previously shown, it is crucial to analyze crowd scenes, consequently, a number of survey articles have been proposed. The survey articles that are now available either mandate the application of conventional computer vision techniques for the examination of crowd situations or focus on a particular facet of crowd analysis, as in crowd counts. [8,9]. This overview article seeks to offer a thorough examination from the development of tools for crowd scene analysis to the latest advancements in deep learning [10,11]. Crowd activity recognition and crowd counting are the two main elements of crowd analysis, which are both included in this survey. Crowd scene analysis is when we look at how many people are in a certain area. This is important for things like surveillance, planning cities, and keeping the public safe. This survey explores different ways of counting crowds, from traditional techniques to the more advanced method called deep learning. This text tells us about different ways to figure out how many people live in different places. It talks about the good and bad things about these methods and what they can teach us [12,13].
How to Cite this Article:
Faisel G. Mohammed, Abbas F. Nori, Noor N. Thamer. A Comprehensive Literature Survey for Crowd Scene Analysis techniques. International Journal of Contemporary Research in Multidisciplinary. 2023: 2(6):16-23
Download PDF
