Unveiling Anomalies in Surveillance Videos Through Various Transfer Learning Models

Video anomaly detection touches upon the process of automatically identifying abnormal events or behaviors in video streams. It plays a crucial role in various domains, including security, surveillance, and safety monitoring. Traditional methods of video analysis often rely on manual inspection, which often requires a significant amount of time and is susceptible to human error. The concept of video anomaly detection aims to overcome these limitations by leveraging advanced technologies such as computer vision and machine learning. This research study performs a relative analysis on the models of deep learning such as VGG16, Resnet50, VGG19, DenseNet121 to deal with video anomaly detection. This study uses subsets of various classes of UCF crime datasets. The UCF Crime Dataset contains 1903 video clips captured from real-world surveillance cameras. Results show Densenet-121 with an ROC-AUC score of 0.85 performs better when vis-á-vis the other models used in this study.