An intelligent unsupervised anomaly detection in videos using inception capsule auto encoder

ABSTRACT

The feature extraction model focuses on extracting the spatio-temporal features using an Inception-CAE. The extracted features have been enriched with essential data to achieve the abnormal events. In addition, the reconstruction error between the initial and the reconstructed video frames is estimated. Finally, the normality score is calculated and related to the threshold calculated using the CTOA. The proposed system is implemented in Python by initiating datasets from CUHK Avenue, UCSD Ped2 and LV. Performances are evaluated in terms of ROC, AUC, accuracy, loss, model size and time complexity. Then the performance of the proposed system is compared to recent existing techniques. The maximum AUC obtained by the proposed system is 99.1% for CUHK Avenue, 99% for UCSD Ped2 and 99.2% for LV datasets. Thus, the simulated outcomes clearly showed that the proposed system had achieved better performance against classical techniques for detecting the anomaly in videos.

KEYWORDS:

• Threshold block difference
• Max_D Shot Type
• Inception-CAE
• coyote threshold optimization algorithm
• anomaly detection
• key frame selection
• video shot segmentation
• unsupervised deep learning

Disclosure statement

No potential conflict of interest was reported by the author(s).

Data availability statement

Data sharing is not applicable to this article.

Additional information

Notes on contributors

Harshadkumar S. Modi

Harshadkumar S. Modi has received his Bachelor of Engineering degree in Computer Engineering from U.V. Patel College of Engineering, Hemachandracharya North Gujarat University, Gujarat, India in 2006, and Master of Engineering degree in Computer Engineering from Government Engineering College, Gandhinagar, Gujarat, India in 2018. Previously, he worked as an Assistant Professor with Information Technology Department, Ganpat University, Gujarat, India. He is currently working as a Lecturer with Computer Engineering Department, Government Polytechnic Gandhinagar, affiliated with Gujarat Technological University Gujarat, India and has a teaching experience of more than 15 years. His research interest includes Computer Vision, Deep learning applications, Crowd Behavior Analysis and Visual Tracking.

Dhaval A. Parikh

Prof. (Dr.) Dhaval A Parikh has completed his bachelor degree from Gujarat University, Gujarat, India in 1991, master degree from Sardar Patel University, Gujarat, India in 2003 and PhD from C. U. Shah University, Gujarat, India in 2017. He has a teaching experience of more than 22 years. Currently he is working as a Professor and Head of the Department with Computer Engineering department, Government Engineering College, Gandhinagar, affiliated with Gujarat Technological University Gujarat, India.