Enhanced Video Monitoring Using Stae Autoencoder

Big data applications have become widespread across different sectors and research domains, incorporating a vast array of sources such as social media information, sensor data, agricultural data, medical data, and data derived from space exploration. A significant contributor to unstructured big data is the use of video feeds from surveillance cameras. These cameras are deployed extensively in critical areas where security is paramount. The manual monitoring of surveillance footage is both labour-intensive and time-consuming. Security concerns in this context vary from identifying thefts to detecting violent incidents and even anticipating potential explosions. In densely populated public spaces, security encompasses monitoring a broad spectrum of abnormal events, with violence detection being particularly challenging due to the complexity of group dynamics. Analysing anomalous activities in crowded video scenes presents substantial challenges due to real-world constraints. This survey paper delves into a comprehensive review starting from action recognition to crowd analysis, culminating in the intricate task of violence detection in crowded environments. The majority of the studies reviewed in this survey rely on deep learning techniques.In this paper, we employ the State Spatiotemporal Auto-Encoder (STAE) alongside a combination of LSTM (Long Short-Term Memory) layers and convolutional networks. These advanced techniques are pivotal in our approach to tackle the challenges of analysing surveillance videos, particularly in detecting anomalous activities such as violence in crowded environments.