This project by Kingfisher was undertaken as a ‘proof of concept’ to locate koalas in commercial eucalypt plantations using Unmanned Aerial Vehicles (UAVs) and ground-based platforms fitted with thermal digital imaging cameras and suitable ‘koala-identification’ software algorithms. This more affordable technology, in combination with ecological knowledge and computer vision algorithms developed by Australian Centre for Field Robotics (ACFR), offers a tangible alternative to manual (human) ground-based assessments.
The study objectives were to:
- Improve speed and accuracy in locating koalas in advance of harvesting.
- Deliver a system based on UAVs that provides considerable savings to the industry, ranging from decreased staff and equipment costs to improved accuracy in population counts.
- Provide precision in locating koalas.
- Provide access to terrain that would be otherwise dangerous or difficult on foot and reduce the associated safety concerns when traversing these areas.
Five trials using various settings, heights and image overlaps were conducted over several months in the Bessiebelle forest area in western Victoria. The data was then processed for the development of an algorithm to allow computer identification of koala locations thus providing geo reference coordinates to forest operators in real time so they could avoid koalas during harvesting.
The ACFR computer algorithm was successful in identifying animals although with some false readings. The algorithm is a continuous learning system and will improve over time with continued data collection and use enabling removal of the incorrectly identified animals.
In addition to the above trial, the Flir camera was tested to see if a suitable ground-based system could be developed. The ground-based application was found to be ineffective.
Overall, the infield operations could be conducted by two personnel and each flight would take less than eight minutes covering around 1 hectare, providing a cost effective solution to koala identification prior to harvesting.
A project summary is also available via this link.
Project number: VNC389-1516
