Abnormal High Density Crowds

The paper presented by Samar Mahmoud; and Yasmine Arafaf et, al a novel dataset called the "Abnormal High-Density Crowd Dataset," addresses the challenge of anomaly detection in crowded environments, particularly focusing on high-density crowds—an area that has received limited exploration in computer vision and crowd behaviour understanding. The dataset is introduced with considerations for privacy, annotation accuracy, and preprocessing.

The proposed dataset aligns with the constraints of benchmark datasets like UCSD, UMN, and Avenue, featuring occurrences of both normal and abnormal behaviours along with validated annotations for abnormal behaviour. Notably, this dataset is unique in that it specifically includes footage of high-density crowds, distinguishing it from benchmark datasets that predominantly cover low- to medium-density crowds.

The paper emphasizes the importance of adhering to privacy considerations, ensuring the accuracy of annotations, and applying appropriate preprocessing steps to enhance the dataset's quality. The evaluation of the dataset involves testing it against state-of-the-art crowd anomaly detection methods. Results indicate that training/testing these methods on high-density crowds decreases detection performance, suggesting the challenges associated with detecting anomalies in densely populated scenarios.

Published in the Fourth International Conference on Multimedia Computing, Networking, and Applications (MCNA) in October 2020, the paper contributes to the field by providing a specialized dataset that fills a gap in existing datasets, enabling researchers to explore and develop more effective crowd anomaly detection methods for high-density environments. The DOI for the paper is 10.1109/MCNA50957.2020.9264277, and the IEEE published it. The conference took place in Valencia, Spain, during the specified dates.

Anomaly detection within crowded environments is a key challenge in the fields of computer vision and crowd behaviour understanding. Furthermore, anomaly detection within high-density crowds remains an insufficiently explored area. In this paper, we propose a novel abnormally high-density crowd dataset. The proposed dataset adheres to the same constraints as some of the benchmark datasets, such as the UCSD, UMN and Avenue datasets.