DEEP LEARNING APPROACH FOR SUSPICIOUS ACTIVITY DETECTION FROM SURVEILLANCE VIDEO

Abstract

The importance of video surveillance in modern society cannot be overstated. Once AI, ML, and
DL were introduced, the technologies were already too far forward. Utilizing the aforementioned
permutations, several methods have been developed to discern between different types of
suspicious behaviour based on the live monitoring of film. Human conduct is the most erratic,
and it's sometimes hard to tell whether an unusual pattern of behaviour is really typical. An alarm
message is sent to the appropriate authority if the system predicts suspect behaviour in a school
setting, and if the activity is normal, no alert is sent. In many monitoring applications, a series of
frames taken from a video are used in rapid succession to conduct checks. The whole structure
may be broken down into two halves. First, a classifier uses video frame computations to acquire
features; next, using those characteristics, a prediction of whether a frame is suspicious or not
may be made.