Earlier this year, the Humane Rescue Alliance along with its partners PetSmart Charities, ASPCA, HSUS, and independent researchers wrapped up a three-year effort to count the number of cats in the District of Columbia. The goal of the DC Cat Count was simple: to estimate the total cat population in our nation’s capital and develop tools and protocols to help inform data-driven cat population management programs that can be applied to other communities.
One technique used to count the free-roaming cats who live outdoors was a camera survey, which used trail cameras that take a picture of anything that moves within their field of view, including cats (and lots of wildlife).
However, processing these pictures to turn them into usable data proved to be time-consuming. For example, DC Cat Count staff, volunteers, and collaborators at Smithsonian Conservation Biology Institute looked through nearly six million photos from more than 1,500 cameras one by one to tag the species of each animal. A promising alternative to this painstaking task would be the use of artificial intelligence (AI) to create and use algorithms that could automatically identify cats in the trail camera images.
This technology would drastically reduce the processing time required for wildlife camera images, making this cat counting method more feasible for others who seek to count the cats in their municipalities. But there has been a missing link preventing development: a diverse image dataset with complete species identification to train the algorithms.
Now, the DC Cat Count’s monumental photo processing effort has filled this gap by creating a uniquely detailed image dataset that can help train machine learning algorithms to identify cats and urban wildlife in camera trap pictures.
To make full use of this opportunity, HRA partnered with Conservation X Labs (CXL), a company developing solutions to prevent extinction, to share the DC Cat Count image dataset and train AI algorithms that can be used to process photos in future cat counts in municipalities nationwide. This partnership has successfully created a cat identification algorithm that can be deployed on CXL’s Sentinel devices.
The Sentinel is CXL’s new technology that retrofits existing environmental sensing devices typically used by conservationists, such as trail cameras and acoustic recorders. The Sentinel makes these tools smarter by giving them the ability to run AI algorithms on data (such as images) as they are captured and send notifications to users in real-time. For example, it allows users to know if something critical, like the presence of a poacher or endangered species, is detected so they can take immediate action.
DC Cat Count’s photos have provided the CXL engineering team the ability to train a model to identify whether cats are present within a photo. In the below images, the model trained was able to pick up when multiple cats were seen in an image and draw a bounded box around the specific locations of each animal. This dataset can be used further to train additional models on specific features of cats, from identifying sex to whether a cat is wearing a collar.
By combining the DC Cat Count’s unparalleled image dataset and CXL’s artificial intelligence expertise, this partnership has increased the accessibility of trail cameras as a cat population estimation and monitoring tool for animal welfare researchers and practitioners.
We'll send fresh, amazing content straight to your inbox so you can keep a pulse on your animal community.