Using UAVs and machine vision in the early detection of combine harvester fires

Mosetter, Steven (2016) Using UAVs and machine vision in the early detection of combine harvester fires. [USQ Project]

Text (Main Project)

Download (9MB) | Preview


Fire during the harvest of crops is an ever present hazard. The combination of hot and dry conditions with a highly flammable crop material creates perfect conditions for fire to start and propagate, the result of which can be loss of production, time, equipment and the crop itself.

The aim of this project is to create a system that can actively detect fire activity so that the harvester operator has a better chance of containing the fire before it spreads out of control. By using the ability of CCD cameras to detect Near Infrared (NIR) and sophisticated machine vision, a cheap and effective fire detection system can be created that can alert the operator to any developing fire before the grows out of control.

An extensive review of available literature regarding combine harvester fires, the use of Near Infrared (NIR) and visual light cameras in fire detection and the use of machine vision to detect fire was conducted. An experimental prototype NIR camera system was constructed with off the shelf components selected on the basis of suitability and cost and a computer program was developed with the purpose of detecting fire in the video feeds.

Testing was done in two phases. The first phase was to test the hardware of the system to determine if the cameras was even able to see fire or related phenomena. The second phase of testing was to determine if the machine vision software was able to quickly and accurately identify fire under different circumstances, and its ability to filter out other phenomena that may cause false positives.

The hardware of the system was able to detect fire in most circumstances. Inexpensive cameras operating in the NIR and Visual spectrums are more than capable of seeing the light, heat and smoke emissions of the fire under all of the conditions that such a system would likely encounter during normal operations. The emissions that the camera detects is highly dependent on the proximity of the camera to the fire, which has significant implications on the software processing algorithm and its ability to accurately detect a fire.

The software algorithm was able to correctly identify a fire during all the software tests. The NIR camera was able to correctly identify fire in all of the testing but was subject to false positives from reflections. The colour program had greater success in bright conditions as the reduced contrast between the fire and its surroundings enabled the colour of the fire to be more easily seen.

Detecting fire with machine vision is still a field that is in its infancy, but the results gained from this project are very promising and with further development could yield a system able to reliably detect fire in harvest conditions.

Statistics for USQ ePrint 31450
Statistics for this ePrint Item
Item Type: USQ Project
Item Status: Live Archive
Additional Information: Bachelor of Engineering (Honours) Major Mechanical Engineering project
Faculty/School / Institute/Centre: Historic - Faculty of Health, Engineering and Sciences - School of Mechanical and Electrical Engineering (1 Jul 2013 - 31 Dec 2021)
Supervisors: Low, Tobias
Date Deposited: 21 Jul 2017 02:03
Last Modified: 21 Jul 2017 02:03
Uncontrolled Keywords: machine vision; UAV; combine harvester; Near Infrared (NIR); CCD cameras
Fields of Research (2008): 09 Engineering > 0913 Mechanical Engineering > 091302 Automation and Control Engineering
Fields of Research (2020): 40 ENGINEERING > 4007 Control engineering, mechatronics and robotics > 400799 Control engineering, mechatronics and robotics not elsewhere classified

Actions (login required)

View Item Archive Repository Staff Only