Abstract:
Security is of paramount importance in today’s world. Public places such as shopping malls, banks, ATMs, city squares, and parks are increasingly equipped with CCTV cameras. These cameras aid in monitoring these spaces and keeping them safe for citizens to use. However, such a large amount of video data cannot be constantly monitored in real-time by humans. Such monitoring would require trained, vigilant workers whose sense of judgement can be trusted. Anomalous behaviour is rare, making the job harder to perform for humans. Additionally, the definition of such behaviour varies by time, place and context. As a result, there is a large demand for this monitoring to be automated. Such automation would need to be accurate, fast and reliable. It would lead to better security and enable monitoring in a larger area. This work aims to use unsupervised deep learning networks to automatically identify anomalies in such data and report them in real-time.
