Preliminary Investigation into Machine-Learned 2D Automated Video Surveillance Systems

Abstract

The purpose of this study is to explore the possibility of using 2D human pose estimation algorithms in automated video surveillance systems. The proposed concept boils down to develop an intelligent application that can handle the incoming results and even prevent the occurrence of critical events in the monitored areas. The so-called critical events represent the acts of vandalism, crime, burglary, theft, fights, etc. The core aspect of the described solution is the use of conventional 2D imaging devices, i.e. CCTV cameras, IP webcams or even mobile phone cameras and an AI (artificial intelligence) neural network trained on a set of images. That makes the proposed solution relatively cheap, universal and easy to install across a wide range of target environments without the need for additional sensors. In the paper the authors highlights the architecture of the system, presents details of the video data acquisition stage followed by a description of the 2D human pose estimation toolkit. It is then complemented by summarizing the data extraction stage from pre-recorded clips for machine learning.