Due to the rapid growth of urbanisation and the rise of industry, there is an increased demand for surveillance system that work in real time. Artificial intelligence-based anomaly identification system only addresses a portion of the difficulties, primarily ignoring the dynamic nature of aberrant or abnormal behaviour across time. Using a training dataset with established normality and known error values are two further drawbacks of anomaly identification systems. A new approach to identify anomaly and its localization of video stream in real time called the Step Incremental Learner (SIL), is presented in this study. As new anomalies and normalities emerge over time, the unsupervised deep learning technique known as SIL uses active learning with fuzzy aggregation to continually update and identify them. Three benchmark datasets are used to show and assess SIL in terms of accuracy, robustness, computing overhead, and contextual indicators. Experiments conducted by our team show that the proposed system is very effective for 24x7 surveillance of video systems.