Efficient Video Anomaly Detection using Residual Variational Autoencoder

Video anomaly detection is a critical task in various fields such as surveillance, security, and transportation, and has been gaining significant attention in recent years.Manually monitoring such anomalies can be time-consuming and monotonous in result. Detecting anomalies in videos is difficult because of erratic nature of the event. Motivated by these issues we purpose a promising approach for video anomaly detection by using Residual Variational Autoencoder(RVAE) model which is able to detect anomalies in an unsupervised manner.RVAE can capture more complex patterns in the data and improve the reconstruction error of the model, In this model, the encoder takes the input and provides a low-dimensional latent representation of it, and the decoder learns to reconstruct the original input with minimum loss. ConvLSTM layer is used to make better Spatio-temporal learning and residual connection to reduce the vanishing gradient problem. This model is implemented on three benchmark datasets ucds pedl, ped2, and Avenue datasets given the result shows good potential and it could be a step forward to improve the performance