Suzanne Roy6
Posted in Problem Bank and visible to The Public
My organization is currently implementing the world's largest fertility control program for wild horses. We're managing a population of 3,000 mustangs on 300,000 acres in northern Nevada. The program utilizes an immunocontraceptive vaccine, PZP, darted remotely to curb reproduction. The foundation of the program is a data base that includes over 3,000 horses, individually identified by color, marking, social affiliation and other identifying features (including which side of the neck the mane falls on.) To identify which horses need to be darted, our field volunteers must search through the database, using key characteristics to sort through the data. This can be time consuming when in the field with a group of horses, who may run off at any moment.

We are looking to apply technology to make the darting process more efficient. We're seeking a modernized app (our database technology is functional but somewhat dated) and horse recognition software that would allow us to identify individual horses based on a photograph, similar to what is being done for zebras, elephants, whales, chimpanzees, etc. Such technology would revolutionize our fertility control program, allow us to further scale it up and establish it as a model for humane management for wild horses populations - a hot button issue today out here in the American West.

The team at WildMe - @Jason Holmberg or @Jason Parham - may be able to help, or at least point you in the right direction. I know there are a number of others in the community who are interested in applying ML tools to real-world conservation challenges.

Question: would you envision that the tool work in situ, where a live camera is pointed at the horses and they are identified in the database on the live camera? Or is it enough to take a photo of the group, upload, and be told which have been darted and which haven't?

I would also say - if there are other solution types that the community could come up with (non ML related) to identify individual horses, I assume they would be welcome?

by Tom Quigley
Thanks Tom. The thought was take a photo, upload and get the id info on the horse, including darting history. Limitation on that would be cell reception. Absolutely open to other solutions.
by Suzanne Roy
Hi Suzanne, Seems like a really important issue and one very much tractable with Machine Learning as Tom has said. With a history of applying machine learning to image classification problems and deploying these for users via (web) apps, I would be keen to express interest in the project. I would certainly be happy to collaborate with others if this aligned with them also. Let me know what you think.
by Ben Harries
Hi Ben - would love to have my Dir. of Field Operations connect with you about this. What is best way to communicate? 

by Suzanne Roy
Great! Best way would be through email:
by Ben Harries
Thanks Ben! We will be in touch.
by Suzanne Roy
This would be a perfect project for Wildbook and it's machine learning pipelines!  We have run across this project in a few places and have been recommended an engagement, which we wouldn't mind sitting down since we are all stuck at home to have a further discussion.  

Wildbook as an ecological data management tool would be the place to store the things like dart history and other sighting records about a given individual.  If a picture is taken of the animal and it provides enough visual information for an ID, we would also begin the work of automating the search and comparisons like we do for other species (e.g. zebras, giraffes, sea turtles, etc.)
by Jason Parham
Couldn't agree more, Wildbook would be a perfect tool to use for this use case. Are you planning on only using standard RGB cameras for detection? Are there other physical features that can distinguish individual horses from one another?
by Henrik Cox
Yes. Wildbook is a leader in this field although to date has not focused on wild horses. Our database has thousands of photographs taken by digital cameras with zoom lenses. We also have data from wildlife cameras set up in different areas. Other identifying marks besides color and markings, could include whorls (patch of hair growing in opposite direction of rest of hair. found on stomach, face, stifle, hocks.) Not sure if face recognition on a horse would work. I believe only app developed so far can tell cows from horses but not identify individual horses. Here is example of entry in our database:
by Suzanne Roy