Title: Automated anomaly detection and multi-label anomaly classification in crowd scenes based on optimal thresholding and deep learning strategy

Authors: Harshadkumar S. Modi; Dhaval A. Parikh

Addresses: Computer Engineering, Gujarat Technological University, Gujarat 382424, India; Computer Engineering, Government Polytechnic Gandhinagar, Gandhinagar, Gujarat 382027, India ' Computer Engineering, Government Engineering College Gandhinagar, Gandhinagar, Gujarat 382027, India

Abstract: The anomaly detection present in the crowd scenes acts as an important role in the automatic video surveillance systems to alert the casualty in the field that suffers from the huge quantity of footfalls. In this paper, new anomaly detection and multi-label anomaly classification are planned in crowd scenes using the enhanced deep learning strategy. A modified deep learning model called enhanced recurrent neural network (E-RNN) is used for the multi-label anomaly classification. As the main contribution to this paper, the threshold for movement score and appearance score and the number of hidden neurons of RNN is tuned or optimised by the hybrid elephant herding-grey wolf optimisation (EH-GWO), which helps to attain the best detection and classification accuracy. The experimental outcomes reveal that the developed deep learning model attains a higher accuracy in comparison with other established approaches on benchmark datasets.

Keywords: automated anomaly detection; multi-label anomaly classification; optimal thresholding; convolutional neural network; E-RNN; enhanced recurrent neural network; EH-GWO; elephant herding-grey wolf optimisation.

DOI: 10.1504/IJAACS.2024.137006

International Journal of Autonomous and Adaptive Communications Systems, 2024 Vol.17 No.2, pp.127 - 158

Received: 27 Oct 2021
Accepted: 01 Feb 2022
Published online: 01 Mar 2024 *

Full-text access for editors Full-text access for subscribers Purchase this article Comment on this article