Sehmish-Lahey, Laura (2018) Identification of Kangaroos using Machine Vision. [USQ Project]
Abstract
Since farming was developed, people have been trying to find ways to protect their crops and animals from pests. As the demand for food grows, the importance of controlling pest populations increases, and new more innovative ways are being found to do this. One such method which has recently become popular in Southern Queensland is the erection of exclusion fences to slow down the movement of target animals. While methods like this seem good in theory, there is no real way to know how effective they are. This project aims to design a machine vision algorithm that can identify and count kangaroos as they cross in front of a camera, with the purpose of providing farmers with a way to monitor the population of kangaroos on their property.
A literature review into the history and methods of using machine vision to identify moving objects revealed many different methods of image and video analysis, as well as the various camera types, that could be used for the project. Before the algorithm could be created, four different locations along a forestry fence were chosen for the collection of footage. These sites, and their footage, were analysed and a single site was chosen to be the basis for the algorithm. The chosen site was located parallel to a boundary fence 3 m away from a 40 cm hole in the netting, as this provided side profile footage of the kangaroos as they left the hole.
The research project used footage from the selected location to design a machine vision algorithm based on frame differentiation and blob detection. OpenCV libraries were used to assist with the image analysis operations, and target animal blobs were analysed based on their size, shape and trajectory. The algorithm was then refined using footage of non-target animals to increase its level of species discrimination. A crossing line was used to track the number of target animal blobs that travelled across the frame. The algorithm was then tested with footage recorded at different times of the day, and at different locations, to provide an indication of its effectiveness. The developed method performed better during the afternoon as it detected 76% of kangaroos and 100% of humans, but only 33% of dogs. It was determined that the algorithm was reasonably successful at correctly identifying 76% of kangaroos at the selected location but was less successful at the other locations due to the varying movement profile of the kangaroos. In addition to this, it was very successful in the rejection of humans as target animals but was less successful in rejecting dogs. The algorithm was also tested against a generic multiple object tracking algorithm, and it performed better at both detection and counting of kangaroos on the test footage.
The algorithm designed in this dissertation leaves room for future research in the kangaroo identification field. Improvements could be made in the algorithm to reduce the detection of dogs in the footage, or to increase the number of species being tracked past the camera. Further work could also be done by creating a machine vision prototype to analyse the footage in the field.
|
Statistics for this ePrint Item |
| Item Type: | USQ Project |
|---|---|
| Item Status: | Live Archive |
| Additional Information: | Bachelor of Engineering (Honours) (Mechatronic) |
| 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; McCarthy, Cheryl |
| Date Deposited: | 05 Sep 2022 04:03 |
| Last Modified: | 05 Sep 2022 04:03 |
| Uncontrolled Keywords: | kangaroos; exclusion fences; machine vision algorithm; identify and track |
| URI: | https://sear.unisq.edu.au/id/eprint/40757 |
Actions (login required)
 |
Archive Repository Staff Only |
