A model for video anomaly detection using deep learning-based Recurrent Neural Networks

The Automatic Detection of Anomalies in Video Surveillance (ADA-VS) is an intriguing field of research. Challenges in identifying unexpected events, such as illegal activity and assaults, persist despite developing multimedia-based Anomaly Detection (AD) algorithms. Problems, including spillage, visual disturbances, irregularities, and goals, make it hard to spot VS anomalies. This paper introduces a new approach to Background Deduction (BD) in VS utilizing a state-of-the-art Recurrent Neural Network (RNN) -based Deep Learning (DL) framework and the Fully Consistent Extremal Area (FCEA) feature extraction methodology. In this research, BD is analyzed using the GMM method. The method inputs a target edge and a sample image that does not include any abnormalities. After making short-term adjustments to these inputs, it generates a segmentation guide with the same spatial goals. The guide highlights the pixels representing noticed anomalies, defined as the elements eliminated in the source image. The proposed ADA-VS system using RNN outperformed the existing methods by achieving improved simulation results.