Suspicious Action Detection and Recognition in Remote Areas Using AI and Machine Learning Techniques

Abstract

The use of video monitoring to detect suspicious behaviors in public transportation zones is gaining popularity. For post-event analysis, such as forensics and riot investigations, automated offline video processing systems have been employed in general. However, there has been relatively little progress in the area of real-time event identification. We present a framework for processing raw video data received from a fixed colour camera set at a specific site and making real-time inferences about the observed activities in this research. First, using a real-time blob matching method, the proposed system gets 3-D object-level information by recognizing and tracking individuals and baggage in the scene. Using object and in turbojet motion characteristics, behaviors and events are semantically detected based on the temporal aspects of these blobs. These supervised machine learning techniques are used to detect and track social distancing between one or more people's movements in public spaces, and these observations can be made using CCTV footage. To show the potential of this technique, a variety of sorts of behavior that are significant to security in public transportation locations have been chosen. Abandoned and stolen items, fighting, fainting, and loitering are examples of these. The experimental findings reported here illustrate the approach's outstanding performance and minimal computing complexity using common public data sets.