Title: Hybrid VGG-DenseNet for anomalous behaviour detection in crowded video scenes

Authors: P.S. Rohini; I. Sowmy

Addresses: Department of Electronics and Communication, Noorul Islam Centre for Higher Education, Kumarakovil, Thuckalay – 629180, Kanyakumari District, Tamil Nadu, India ' Department of Biomedical Engineering, Noorul Islam Centre for Higher Education, Kumarakovil, Thuckalay – 629180, Kanyakumari District, Tamil Nadu, India

Abstract: With growing interest in detecting suspicious or abnormal behaviour in crowds, the field of crowd anomaly detection is growing fast. Using a deep learning approach, this research aims to implement an improved model for anomaly detection. Phases such as pre-processing, object detection, feature extraction, and detection are covered by the proposed approach. Bilateral filtering is used to pre-process the video frames after they have been initially extracted from the videos. Following that, the convolutional neural network (CNN) model is used to carry out object detection. Subsequently, colour, shape, and texture features are extracted as part of the feature extraction process. Lastly, the suggested visual geometry group (VGG)-DenseNet algorithm, which combines the VGG and DenseNet algorithms, is employed for anomaly identification. Experimental outcomes states that presented mechanism attained an accuracy of 95.4%, sensitivity of 97.5%, specificity of 89.5%, precision of 96.1%, FNR of 0.024, FPR of 0.104, and F-measure of 96.8%.

Keywords: anomaly behaviour detection; deep learning; neural networks; object detection; feature extraction.

DOI: 10.1504/IJSN.2024.143768

International Journal of Security and Networks, 2024 Vol.19 No.4, pp.199 - 209

Received: 25 Jun 2024
Accepted: 27 Oct 2024
Published online: 06 Jan 2025 *

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