Anomaly Identification in Surveillance Video Using Regressive Bidirectional LSTM with Hyperparameter Optimization
Rajendran Shankar, Narayanan Ganesh · 2024
Urban planners and academics are influenced by the idea of smart cities to develop sustainable, modern, and reliable infrastructure that offers their citizens a respectable standard of living. To meet this demand, there have been video monitoring devices installed to improve public security and welfare. Despite scientific advancements, it is difficult and labor-intensive to identify odd events in surveillance video systems. In this research, we concentrate on the improvement of anomaly detection in intelligent video surveillance using regressive bidirectional Long Short-Term Memory (LSTM) (RBLSTM) with hyperparameter optimization (HO). The suggested framework is tested on a real-time dataset, the ShanghaiTech Campus dataset, and it outperforms state-of-the-art techniques in terms of performance. It is important to take advantage of higher-quality features from accessible videos. This work uses the Video Swin Transformer model to extract features. As a consequence, anomaly detection in video surveillance applications provides reliable outcomes for real-time situations. In this study, an abnormality was correctly identified in videos with a 98.5% accuracy rate. Future research might explore further experiments by investigating other methods to reduce the noise present in the positive bag.