This project aims to progress the field of conservation technology by developing face recognition software for use in wildlife monitoring. Using human face recognition techniques, we are developing a software tool that can identify individual brown bears (Ursus arctos) from images of their faces. Applying this technology to camera trap imagery would provide scientists with a new technique to monitor wild populations of brown bears and ask a wider variety of applied research questions. This is important as scientists are under increasing pressure to draw larger conclusions from their research, but with fewer resources available. In addition, we plan to test the software in the field and develop guidelines for its use. This project provides the foundation for the development of face recognition for other threatened wildlife, which could aid conservation efforts worldwide.
Melanie Clapham is a Director and Conservation Scientist for the BearID Project. She holds a PhD in Conservation Biology from Lancaster University in the UK
Ed Miller is a Director and Software Developer for the BearID Project. He has 25+ years of experience developing hardware, software and systems.
Mary Nguyen is a Director and Software Developer for the BearID Project. Mary has 25+ years of experience in software development.
Deep learning is a form of machine learning, typically employing large neural networks to learn data representations from training data. Since 2012, all the winners of the premier computer vision contest, ImageNet Large Scale Visual Recognition Challenge, have utilized deep learning. Machine learning is already being applied to numerous wildlife classification and recognition problems.
Camera traps have revolutionised the way we observe and study wildlife. From citizen science census projects to investigations into behavioural ecology, camera traps allow us to observe wildlife in a way not previously possible. Camera traps belong to a class of wildlife monitoring techniques referred to as ‘non-invasive’. They allow for data collection without humans being present and therefore are often considered as less stressful to individuals.
One challenge of using camera traps to study bears is the inability to consistently recognise individuals, due to the lack of unique natural markings for some species. Methods have been developed to try to account for a lack of individual identification in population inventory studies using cameras, but concerns over reliability remain. Automated methods to detect and identify both ‘marked’ and ‘unmarked’ wildlife in images and video footage are starting to receive increased interest.
Object detection with minimal code using Microsoft Azure Custom Vision SDK.
Creating an image classifier without writing any code using Microsoft Azure Custom Vision.
