AWS Builder Center
Building for Good: AWS Hero Ed Miller and the BearID Project

Building for Good: AWS Hero Ed Miller and the BearID Project

An AWS Hero used facial recognition to help wildlife researchers identify individual bears. This is the first in our Building for Good series, spotlighting community members using tech to solve real-world problems.

Series: Building for Good: Community Stories (2 articles)

  1. 1
    Building for Good: AWS Hero Ed Miller and the BearID Project This article
Sr Program Manager
It started with a webcam and a bear that looked a little different...
A few years ago,  AWS Hero   Ed Miller  was doing two things at once. He was watching  brown bears fish for salmon on a live webcam  in Alaska, and taking online courses in machine learning. He didn't plan for those two interests to collide, but they did. 
He wanted a project to make the learning stick, something real to build instead of just working through a tutorial. At the same time, he was starting to recognize individual bears on the cam by their faces, learning to tell them apart the same way the seasoned viewers in the chat could. The two things clicked together: what if he could teach a computer to do that too? 
What started as a personal learning experiment has grown into the  BearID Project  , an open-source effort to use AI for non-invasive wildlife monitoring.  

The Problem  

Wildlife researchers around the world use camera traps to study animals. They place cameras in the wilderness for weeks or months at a time, collect SD cards full of footage, bring them back to the lab, and then manually review it all. Thousands of hours of video. Most of the footage is a mix of empty forests, blowing leaves, and shadows. The actual footage with (the correct) animals in it might be 5 to 10 percent of the total time. 
For bear researchers specifically, the work goes even further. Once they find footage with bears, they need to identify individuals. Am I seeing the same bear 10 times, or are there 10 different bears? Understanding that distinction is how researchers track populations, study behavior, evaluate the ecological health of their environment, and map how far animals travel. 
Traditionally, the way to track individual bears has been to trap them, tranquilize them, and attach a radio collar or satellite tag. It's dangerous for both, the animals and humans. Camera trapping offers a non-invasive alternative. But without a way to efficiently process all that footage, it doesn't scale.  

Finding the Right People 

Ed didn't work in wildlife research. He's a software engineer with a day job at  Arm , where he leads technical engagements around machine learning and IoT. But he had the technical skills, interest, and drive to try something. So he joined  WILDLABS , an online community that connects conservation researchers with technologists. 
Within hours of posting that he was working on AI for bears, the community manager connected him with conservation scientist  Dr. Melanie Claphman  in Canada who was approaching the same problem from the research side. Within weeks, they formed the  BearID Project . Ed's wife, Mary Nguyen, also a software engineer, joined as a co-developer. 

Making it Real 

The team's first challenge was data. AI needs labeled examples to learn from, and there aren't stockpiles of neatly organized bear face photos sitting around. Ed had scraped what he could from the web, but partnering with a researcher who had years of field data changed the game. They could actually train a model to tell bears apart. 
But as Ed spent more time with researchers, he noticed something. The individual identification problem, hard as it was, wasn't where they were losing most of their time. It was everything before that. The sorting. The reviewing. The weeks spent just getting through footage to find out what was on it. AI models exist that can classify this kind of data, but they're built for people who are comfortable at a command line. Most wildlife biologists are not. 
So, the project grew. Ed started collaborating on  Trapper , an open-source camera trap management platform, helping port it to run on everything from cloud servers down to Raspberry Pis and Nvidia Jetsons. The goal is to put real, usable tools in front of researchers who need them, not just publish code to a repository and hope someone figures it out. He's documented that work in his  Trapper Keeper blog . 
There's another piece to this that surprised him. Researchers in the field use a term borrowed from fisheries: bycatch. A camera trap set up for bears captures everything else that walks by. Deer, birds, other species. For a bear researcher, that's noise. But for someone studying the broader ecosystem, it's gold. Right now, most of it gets thrown away because no one has time to process it. Better tooling could change that. 

When People Know the Bear's Name 

Meanwhile, back on the bear cams, something else was happening.  Explore.org  hosts the live wildlife cameras that got Ed interested in the first place. They've found that viewers who learn to recognize individual bears become significantly more engaged in wanting to protect them. It's no longer abstract. It's not "save the bears." It's save Otis. Save Grazer. When people know an animal by name and follow its life across seasons, they care in a different way. 
Ed built a  BearCam Companion  web application to help the viewing community figure out which bear they're watching in real time. He's been in conversation with explore.org about integrating it directly into their platform, so the identification happens right there while people are watching. 
The technology has also gone international. Ed brought BearID to Ecuador on a sabbatical, working with researchers there to apply it to  Andean bear conservation . 

Showing Up 

Ed doesn't do presentations for a living. He'll tell you he doesn't love it. But he forces himself because if he doesn't show up and talk about this work, not many other people will. He's presented at AWS re:Invent. He ran a workshop at the first  International Conservation Technology Conference  in Lima, Peru for about 30 to 40 practitioners. 
None of this is funded. It doesn't pay. He fits it in around a full-time job and a family. What keeps him going is pretty simple: he loves wildlife, and there is something deeply satisfying about seeing your work used in the wild.  

Get Involved 

Ed's advice for anyone who wants to apply their technical skills to conservation: find your passion and look for the people who need what you know. 
Whatever you care about, even if it has nothing to do with your day job, there's probably a community somewhere that needs help solving a technical problem. For wildlife specifically,  WILDLABS  is the place. They have channels for acoustic monitoring of birds, ultrasonic monitoring of bats, video monitoring, underwater fishery work, and more. They are always looking for people who can build things. 
"Find your passion and use it to keep you driven, even on those days when your job is taking it out of you," Ed says. "It feels like extra work, but it's also very invigorating when you accomplish something and see it used in the wild." 

Learn More 

Want to learn more? These sites are great starting points:  

Series: Building for Good: Community Stories (2 articles)

  1. 1
    Building for Good: AWS Hero Ed Miller and the BearID Project This article
Any opinions in this article are those of the individual author and may not reflect the opinions of AWS.
Enjoyed reading this content? Let the author know!

Your likes, comments, shares, and saves help creators reach more builders.

Loading recommendations

Loading article