Cloud-based wildlife detection

For one of our clients, we have developed an AI-based system to detect wildlife species using images and videos from trail cameras.

Add-on for bestselling product

HYMA Skog & Trädgård, a company operating in the forestry, gardening and outdoor industry, was in need of an add-on for one of their best-selling products. The product in question is a trail camera – a type of surveillance camera used on hunting grounds and in other outdoor spaces for the landowner to be able to capture (on image, that is) and identify what kind of species are moving in the area, whether it is a deer, squirrel or a disoriented human being. The camera is connected to a cloud service meaning all photos are continuously saved to the cloud.

HYMA wished to extend the product offering to include detecting what kind of creatures images taken by the trail cameras contain. Or how many animals there are. Or maybe if it even is an animal or just a crooked tree… Anyway, you get the issue: our client needed a high-tech solution that could identify the specific animal in the images captured by the camera.

AI solution for animal detection

To be able to identify what kind of animal has been captured by a trail camera, we built an AI-based system which uses machine learning algorithms that retrain based on user feedback. The system is coded in Python and uses TensorFlow and Keras to train a classifier model for inferring the species in images and videos. The solution is deployed in Azure with Kubernetes.

The solution enables our client’s end-users to log onto the cloud and see identified animals in their pictures, as well as give feedback on a classification. This feedback is stored and used for transfer learning, ensuring continuous improvements in the solution’s machine learning algorithm.

An even more high-tech product

For our client, a snazzy high-tech feature has been added to their already great product. And for their users, where previously the camera’s owner had to focus – eyes squinting, almost feeling the headache coming – when trying to distinguish what animal the camera had captured, it is now done automatically.

In terms of numbers:

  • 156.000 animals are currently identified per day
  • 437 manual corrections are made by users per day
Our best-selling trail camera already offered some rather high-tech solutions for the users. For example, they can change the settings remotely and as well as access all images in the cloud. With this AI add-on, our users now get an even better experience that makes the detection process a lot easier.

                       Anders Carlsson, CEO at HYMA Skog & Trädgård