Videobehavior possible identification and recognition of abnormalities and normal behavior profiling for anomaly detection using cnn model | [Journal of Theoretical and Applied Information Technology • 2022]

Author(s):

1. VENU MAJJI: ASG Technologies, GITAM (Deemed to be University) Visakhapatnam

2. VANITHA KAKOLLU: Department of CS, GITAM (Deemed to be University) Visakhapatnam

3. MPJ SANTOSH KUMAR: Department of CS, GITAM (Deemed to be University) Visakhapatnam

4. K NAGA SOUJANYA: Department of CSE, GITAM (Deemed to be University) Visakhapatnam

5. B. KANTHAMMA: Department of EECE, GITAM (Deemed to be University) Visakhapatnam

6. G. BABU RAO: Department of CS, GITAM (Deemed to be University) Visakhapatnam

Abstract:

The aim of this Paper is to unravel the matter of modeling video behavior recorded in surveillance videos to be used in online normal behavior recognition and anomaly detection applications. With non-manual marking of the training data collection, a replacement architecture is made for automated behavior profiling and online anomaly sampling/detection. The subsequent are the core components of the framework supported discrete scene event detection, a compact and efficient behavior representation method is developed. Modeling each pattern employing a Dynamic Bayesian Network is employed to gauge the similarities between behavior patterns (DBN). A completely unique spectral clustering algorithm supported based on unsupervised model selection and have selection on the eigen vectors of a normalized affinity matrix is employed to get then actual grouping of behavior patterns. To detect abnormal behavior, a runtime accumulative anomaly measure is implemented, while normal behavior patterns are recognized when adequate visual evidence is out there supported a web survey. This enables the fastest possible identification and recognition of abnormalities and normal behavior. Experiments with noisy and broken data sets gathered from both indoor and outdoor monitoring scenarios show the efficacy and robustness of our approach. It's is demonstrated that in detecting anomaly from an unseen video, a behavior model trained with an unlabeled data set out performs those trained with an equivalent but labeled dataset.

Page(s): 5100-5106

DOI: DOI not available

Published: Journal: Journal of Theoretical and Applied Information Technology, Volume: 100, Issue: 14, Year: 2022

Keywords:

Anomaly , CNN , Dynamic Bayesian Network , Adaptive Video Conversion

References:

References are not available for this document.

Citations

Citations are not available for this document.

Citations

Downloads

Views