The imapct of temporal features in anomaly detection in smart cities

This study aims to build and compare some anomaly detection models based on the Artificial Intelligence and internet of Things to be able to predict the anomaly and detect it. These models are based on several studies on the pedestrian data set UCSD, to provide the best parameters and methods to detect anomaly in smart cities using surveillance cameras, evaluating the accuracy of the models over the speed of the anomaly detection, in order to create an accurate model, also this model is able to extract Spatio-temporal features and uses them as more information about the scenes to get even better. Moreover, we used enhanced methods to evaluate the model as r Reconstruction Error (RRE) and area under curve AUC to understand the uses with the models and be able to enhance them more. We selected the CAE model as we got the best accuracy on our data compared to the other models but still far from the state of art accuracy so we need to create a new hybrid model that surpasses all the models created or to extract even more features that will make our model more performing.