01.06.2017 Views

UWE Bristol Engineering showcase 2015

Create successful ePaper yourself

Turn your PDF publications into a flip-book with our unique Google optimized e-Paper software.

Adam Drake<br />

BEng(Hons) Robotics<br />

Project Supervisor<br />

Professor Melvyn Smith<br />

Identification of Cattle using Computer Vision Biometrics<br />

Introduction<br />

Computer vision biometrics is a vast research topic<br />

with many applications across an array of<br />

industries. This investigation is in the area of<br />

animal biometrics. Currently, the majority of<br />

industrial animal identification tasks are<br />

completed using a form of tagging, involving little<br />

or no computer vision.<br />

This investigation examined current<br />

methodologies for computer vision biometrics, in<br />

both human and animal applications, and<br />

proposed a novel non-invasive method for the<br />

identification of cattle. This investigation was part<br />

of an ongoing project, the ‘Precision Health<br />

Monitoring of Cattle’ by the Centre for Machine<br />

Vision at the <strong>Bristol</strong> Robotics Laboratory.<br />

Core Concept<br />

The central idea to this investigation was that<br />

hidden information contained within the animal<br />

can be extracted.<br />

Coat patterns were examined for their<br />

‘uniqueness’ and it was thought that this would be<br />

sufficient to describe an individual animal.<br />

However, this approach was not robust enough<br />

and suffered from problems such as different<br />

lighting conditions and all-brown cattle displaying<br />

no visible coat patterns.<br />

The selected approach uses the actual geometry<br />

of the animal to better satisfy the requirements of<br />

the system to be robust enough to work with<br />

cattle which have no discernible coat pattern.<br />

Image Processing<br />

In order to extract the required information, the<br />

depth images had to be processed to remove the<br />

background and additional unwanted details.<br />

Then through a series of further image processing<br />

steps, including a Gaussian low-pass filter, the<br />

important parts of the depth image are revealed.<br />

Benefits over Radio Frequency ID Tags<br />

RFID tags in use for cattle identification have<br />

several drawbacks. Firstly they have to be affixed<br />

to the animal in some way, taking considerable<br />

time and effort. Secondly, they suffer from<br />

interference, due to the proliferation of multiple<br />

standards and frequencies, leading to instances of<br />

false-reading and inaccuracy.<br />

Applications<br />

This new method of cattle identification has the<br />

potential to offer increased biosecurity on farms,<br />

reduced labour and time costs associated with<br />

current identification methods and improved<br />

accuracy compared to traditional methods.<br />

Project summary<br />

The investigation presents a review of the<br />

current state of cattle identification and<br />

proposes a novel method for identifying cattle<br />

based on their physical features.<br />

Project Objectives<br />

• Develop a method of identifying cattle<br />

based upon some physical characteristic,<br />

using 3D depth images obtained from a<br />

dairy farm environment.<br />

• Research methods of biometric feature<br />

extraction.<br />

• Perform image processing on images of<br />

cattle.<br />

• Research current computer vision<br />

biometric techniques.<br />

Project Conclusion<br />

The proposed new method of identifying<br />

cattle based upon their physical<br />

characteristics performed well and was able<br />

to extract a set of features for all animals in<br />

the dataset. Recommendations were made<br />

for how to integrate the system into other<br />

applications.<br />

It is proposed that this approach not be limited to<br />

cattle and could be extended to work with many<br />

different animals in the future. There is also the<br />

potential to integrate this method of biometric<br />

feature extraction with an artificial intelligence<br />

algorithm, to further enhance its capabilities.

Hooray! Your file is uploaded and ready to be published.

Saved successfully!

Ooh no, something went wrong!