Unmanned aerial vehicles (UAVs) and artificial intelligence revolutionising wildlife monitoring and conservation
Gonzalez, LF, Montes, GA, Puig, E, Johnson, S, Mengersen, K & Gaston, KJ 2016, Sensors, vol. 16, no. 97, s16010097.
This report describes a technique for automated wildlife detection using Unmanned Aerial Vehicle (UAV) imagery that overcomes many of the challenges associated with ground-based monitoring techniques. In a trial of the technique, a system comprising an UAV equipped with thermal imagery sensors and video processing capabilities detected the presence and locations of koalas in a natural environment with remarkable accuracy to generate a reliable population estimate of the surveyed area.
The authors of this report detail and implement a system for detecting animals of interest using an UAV. As the UAV traverses a survey area, an algorithm is applied to the imagery collected on the ground control station computer that enables the counting, tracking and classification of target wildlife. The Template Matching Binary Mask (TMBM) algorithm searches for images in each frame of video recorded that match provided reference image templates of the target animal. The accuracy of detection is improved if several reference image templates are provided that reflect the full spectrum of morphological characteristics of the species of interest. After a potential match is detected, the sighting is confirmed if the animal’s heat signature is detected via thermal imagery. The koala, given its threatened status in Queensland and sedentary nature, was an ideal focal species for trialling this system. A koala rehabilitation enclosure on the Sunshine Coast with a known number of resident koalas was selected as the study area. At a flight height of 20 - 60 metres, detection of koalas using the TMBM algorithm was 100% accurate as confirmed by ground-truthing surveys. The time taken to detect the koala from the first sighting to confirmation ranged from 1.3 to 2.1 seconds depending on the altitude of the imagery.
Reliable population estimates are important for the effective management of threatened and invasive species. Ground-based monitoring techniques for obtaining population estimates are commonly time and resource intensive. Furthermore, ecological challenges associated with ground-based surveys such as a species’ large geographic range, low population density, inaccessible habitat or cryptic behaviour can hinder the accuracy of and ability to validate the results of these techniques. UAV and artificial intelligence technologies such as those demonstrated here can overcome many of these challenges. These technologies can reduce the time- and resource-intensity of surveying large areas, allow close monitoring of animals with minimal disturbance, and provide new angles for observation that can facilitate animal detection in complex habitats.
Information obtained using systems such as that described in this report can be applied in koala conservation programs for population estimation, pest detection, monitoring of relocated animals, rescue operations, and as a precursor to development approvals or construction in koala habitat areas. Ongoing challenges to the use of these technologies for conservation applications include UAV regulations, operational costs, and issues of public perception.
Summarised by Joanna Horsfall
Disclaimer: The summary of this report is provided for reference purposes only and does not represent the findings or opinions contained in the original report. Although every effort has been made to bring forward the main elements of the report, this review is no substitute for the full the report itself. Should you have any concerns or perceive any errors please contact us and we shall endeavour to rectify and improve the review.