Person detection using an omnidirectional camera to shut off the UV-C light on disinfection robot for individuals' safety

No Thumbnail Available

URL

Journal Title

Journal ISSN

Volume Title

Sähkötekniikan korkeakoulu | Master's thesis

Date

2024-01-22

Department

Major/Subject

Autonomous Systems

Mcode

ELEC3055

Degree programme

Master's Programme in ICT Innovation

Language

en

Pages

56

Series

Abstract

One of the most significant global challenges faced during the COVID-19 pandemic was the need for efficient disinfection methods to prevent the spread of the virus. Ultraviolet-C (UV-C) disinfection robots became increasingly popular due to their ability to kill harmful bacteria and viruses using UV-C light. However, UV-C light can also be harmful to humans, so it is important to ensure that people are not exposed to it during disinfection operations. Existing object detection algorithms are limited to conventional images. These images can be stitched together to get a 360-degree view of surroundings but it can be computationally expensive and they may have different resolutions, lighting conditions, and noise levels. This thesis investigates the feasibility of using motion detection and YOLO (You Only Look Once) object detection to accurately detect individuals present within the disinfection environment using omnidirectional cameras. To achieve this goal, motion detection is implemented and analyzed as well as different YOLO models are trained on the 360 Indoor-dataset. These models are then evaluated on their ability to detect people in unseen environments. The results show that motion detection is not a feasible solution for our application because of its static background requirement, while YOLO demonstrates effectiveness with an omnidirectional camera achieving a mean Average Precision (mAP) of 75.34\%. The results also show that among all the YOLO models YOLOv5m achieved the best overall performance, with a precision of 0.80, a recall of 0.68, mAP of 0.73, and an inference speed of 8.2 ms. This indicates that the YOLOv5m model is a promising choice for person detection in real-time disinfection applications. In addition to the experimental results, this thesis also provides a discussion about how to use our trained YOLOv5m in the industry setting to avoid false detections. Overall, this thesis makes a notable contribution by evaluating motion detection for the disinfection robot and by demonstrating the feasibility of using YOLO with an omnidirectional camera through training on 360-degree images. Furthermore, the thesis identifies a more efficient version of YOLO for industrial deployment, with the potential for further development and improved performance.

Description

Supervisor

Kucner, Tomasz

Thesis advisor

Höglund, Thomas

Keywords

YOLO, person detection, motion detection, omnidirectional camera, 360-degree images, disinfection robot

Other note

Citation