Semi-Autonomous Drones in Wildfire Detection: a proof of concept
Vairio, Tuure (2023)
Julkaisun pysyvä osoite on
https://urn.fi/URN:NBN:fi:amk-2023060521398
https://urn.fi/URN:NBN:fi:amk-2023060521398
Tiivistelmä
The scope of this thesis was to create a proof of concept solution of a semi-autonomous drone swarm that can detect and monitor wildfires. This thesis was ordered by the VTT Technical Research Centre of Finland as a part of their FireMan project. This project tries to create innovative solutions to combat the ever-increasing wildfire thread caused by global warming of the climate.
The main theories used in this thesis were concepts of artificial neural networks and how image detection works, how to train an image detection algorithm, and the basics of aerial searches and patterns that the drone could utilize to detect fire from a selected area.
Used data for the flame detection model was found on the internet, mainly using Google image search and the training algorithm used was YOLOv8. Drones were simulated by using Parrot Sphinx simulator software and the command codes were written in Python with Parrot Olympe SDK. Additional libraries were also used for detection streaming and calculations. Olympe SDK allows the use of physical drones if field tests are to be conducted.
Key findings indicate that a simple image detection model could be used for detecting flames from a video stream sent by a drone. A drone can also be flown by using a computer that sends command signals to the drone. The created proof of concept could be developed into a concrete product with a drone swarm capability. However, there are a variety of subjects to be tackled for this to be a working product, such as swarm command and emergency capabilities.
The main theories used in this thesis were concepts of artificial neural networks and how image detection works, how to train an image detection algorithm, and the basics of aerial searches and patterns that the drone could utilize to detect fire from a selected area.
Used data for the flame detection model was found on the internet, mainly using Google image search and the training algorithm used was YOLOv8. Drones were simulated by using Parrot Sphinx simulator software and the command codes were written in Python with Parrot Olympe SDK. Additional libraries were also used for detection streaming and calculations. Olympe SDK allows the use of physical drones if field tests are to be conducted.
Key findings indicate that a simple image detection model could be used for detecting flames from a video stream sent by a drone. A drone can also be flown by using a computer that sends command signals to the drone. The created proof of concept could be developed into a concrete product with a drone swarm capability. However, there are a variety of subjects to be tackled for this to be a working product, such as swarm command and emergency capabilities.