Learning Models in Crowd Analysis: A Review

Abstract

Crowd detection and counting are important tasks in several applications of crowd analysis including traffic management, public safety and event planning. Automatic crowd counting using images and videos is an intriguing but complex issue that has generated considerable interest in computer vision. During the past several years, various learning models have been developed by considering several factors such as model design, input pathways, learning paradigms, computing complexity and accuracy that increases cutting-edge performance. In this work, the most critical advances in the crowd analysis field are reviewed methodically and thoroughly. Numerous crowd counting models have been arranged according to how well these models perform on different datasets using various learning approaches and evaluation metrics like mean average error and mean square error. This work provides insight into the effectiveness of different learning models for crowd analysis. It will be helpful for researchers and practitioners in choosing the appropriate model for their specific applications.