Hybrid imaging-AI approach for handling critical situations in a fast-changing environment: preliminary study

Adam Surówka,T Tracz,K Mróz,T Zdeb

doi:10.1051/matecconf/202032201048

Adam Surówka, T Tracz + Show 2 more

Open Access

https://doi.org/10.1051/matecconf/202032201048

Copy DOI

Abstract

The purpose of this study is to explore the possibility of using selected imaging technologies in automated video surveillance systems. The main goal of this project is to handle events that may lead to security risks, injuries, etc in various environments without relaying on more conventional sensors such as infrared photocells. For this purpose it is necessary to perform a thorough analysis of the events to be interpreted as situations of interest. It is also important to consider the hardware requirements and restrictions for developing such system. The project requires defining a hardware as well as software platform(s) and their integration into an automated tool. This paper describes the implementation of the famous Microsoft Kinect 2.0 depth sensor (well known in gaming and recreational applications) for shape/skeleton detection, and its integration into an artificial intelligence based platform utilizing selected machine learning methods. The author reveals the system implementation details, and then demonstrates its shape detection capabilities while in operation.

Highlights

The use of vision techniques in surveillance systems of industrial facilities and of public use began in the 1940s
Systems based on Human activity recognition (HAR) enable among others the implementation of tasks related to recognizing life threatening situations [4], preventing crime and vandalism [5, 6], supervision of the sick and elderly [7], biometric face identification [8,9,10,11] and analysis and classification of all forms of human activity that may be of interest in a given situation [1217]
It was found that the developed automated video surveillance system correctly implements the project assumptions

Summary

Introduction

The use of vision techniques in surveillance systems of industrial facilities and of public use began in the 1940s. These systems were called Closed-Circuit Television (CCTV) [1, 2]. Attempts to develop effective recognition of human activities and behavior based on the image from CCTV monitoring have been ongoing since 1980. Human activity recognition (HAR) has revolutionized the area of computer vision research in a wide spectrum of applications [3].

Objectives

Methods

Conclusion