The Process for Detecting Suspicious Activity from Surveillance Video Using Deep Learning

Abstract

Video surveillance is essential in the environment we live in today. When artificial intelligence, machine literacy, and deep literacy are introduced into the system, the technologies are too advanced. Different algorithms are in place that help to distinguish colourful questionable acts from real shadowing of footage using the aforementioned combinations. All locations where security is a top priority have CCTV cameras installed. Manual surveillance sounds time-consuming and tedious. In diverse circumstances, such as that involving explosion risk, violence detection, and theft identification, security can be characterised in a variety of ways. Security in crowded public areas refers to practically all types of aberrant incidents. Since it involves group activity, violence detection is one of those that can be challenging to manage. The most changeable one is a mortal gesture and it's veritably delicate to find whether it's suspicious or normal. The deep literacy approach is used to identify either normal or suspicious activity on academic grounds, and it alerts the appropriate authority if it suspects suspicious activity. Frequently, monitoring is carried out using succeeding frames that have been removed from the recording. There are two corridors running the length of the frame. The features are calculated from videotape frames in the first section, and the classifier predicts the classification as suspicious or normal based on the features obtained in the second section.