Facial Recognitions and Bears (Part 2/3 of Robotics and AI Mini-series)

Today I want to write about one of the more interesting uses of AI and Robotics that I have seen in a while. I will be talking about the research and development that Dr. Melanie Clapham, Ed Miller, and Mary Nguyen conducted to created BearID, and the open source program that uses Face ID to detect and recognize grizzly bears in British Columbia.

Face ID is a technology we see everywhere right now. It is on plenty of personal cellphones, laptops, and is also used on a law enforcement level. The technology maps out an individuals face and learns to recognize certain features or landmarks to repeatably recognize them. As mentioned it is used for security on personal devices but the technology is also being trained to work on an industrial level. The Royal Canadian Mounted Police have made statements about using Face ID to try to rescue people from trafficking rings. But how does this bring us to bears?

In a statement to the New York Times Kikaxklalagee / Dallas Smith, president of the Nanwakolas Council, said “when we started doing land use planning [15 years ago], there was just one provincial bear health expert for the whole province.” The difficulties in working with and around the natural ecosystem in British Columbia were amplified because of the lack of knowledge of the animals and their environment. Tagging bears, much like tagging any other animal such as a shark, has been done for a long time and is a partial solution to this problem but the process raises some issues, it can be difficult and stressful for the animal, and/or hard to accomplish. This is where the three researchers from the University of Victoria enter.

Over the course of three years, Dr. Melanie Clapham, a behavioural ecologist and postdoctoral student, surveilled and captured photos of Grizzly Bears at Knight Inlet, which is on the B.C. coast. Instead of using traditional documentation methods, such as inspecting younger bears during hibernation, Clapham used secret cameras to collect data on bears in the area. After collecting thousands of images from bears in her area she partnered with two software developers and a research center in Alaska to develop an AI technology from her data. Her team’s method to initiate the project was also very interesting. They used a basic framework from human facial detection software and an open source program that is called “Hipsterize Your Dog” which was basically a Snapchat filter to put glass and a mustache on your favourite pet. From this they managed to get BearID to an accuracy of 84 percent, and Clapham has commented that the 16 percent the software gets wrong she would also get wrong while observing the bears.

The 84 per cent accuracy they achieved was mainly attributed to deep learning and unsupervised training which is the process where they feed a neural network training data and see how it then performs on a set of new testing data, and then slowly tweaking the system until they get ideal results that can recognize certain bears. Alexander Loos, a research engineer at the Fraunhofer Institute for Digital Media Technology, did have words of caution over this deep learning method; “The network itself extracts the features, which is a huge advantage, but it’s basically a black box. You don’t know what it’s doing.” As for example, certain variables such as lighting or a certain camera’s white balance could cause the system to associate certain colours with certain bears leading to misclassification.

This technology is a fairly new riff on a well established technology that many use in their daily lives. It still has a long way to go but in it’s early days it shows so much promise in wildlife protection, restorations, and awareness. It also brings up so many options for uses. We could use drones with this to gain a better understanding of some species. We could could track diseases in wildlife populations. We could track how climate change is changing the migratory and hibernation habits of certain species. This is a way of continuing to research, respect, and work around the environment while having an extremely low footprint.

Alex Elliott is a Student Ambassador in the Inspirit AI Student Ambassadors Program. Inspirit AI is a pre-collegiate enrichment program that exposes curious high school students globally to AI through live online classes. Learn more at https://www.inspiritai.com/.

Sources: https://www.cbc.ca/news/canada/british-columbia/grizzly-bear-facial-recognition-software-1.5797525, http://bearresearch.org/, https://www.smithsonianmag.com/smart-news/new-i-offers-facial-recognition-grizzly-bears-180976276/, https://www.nytimes.com/2020/11/11/science/bears-facial-recognition.html,