This website is currently under construction, hence why it might still look and feel a bit rough in certain places.


Design and Build | Digital

AI-powered Romania bear repellers

Using computer vision to keep bears and humans safe.

Whilst working for Q42‘s tech-for-good venture Hack the Planet, I got involved with developing a system to prevent/mitigate human-bear conflict in the Carpathian mountains in Romania. When a bear gets used to humans, they become very hard to deter and become a significant risk for humans and livestock. The only solution then is to either try to capture and re-locate the bear, or to kill it. Either way, it is better to prevent bears from getting used to humans to begin with, and that’s where this project comes in.

Over the course of half a year, my colleagues and I got to work designing a system consisting of 4G connected motion triggered cameras and “repellers”. A repeller is a device that features a loudspeaker and a bright strobe light. Whenever it gets activated, it will start making noise and flashing its light. The system works as follows:

  1. A camera senses motion, takes a picture and sends the picture to our cloud-based system over 4G.
  2. The system uses an AI computer vision model to analyze the image for the presence of bears.
  3. If a bear is detected, an activation trigger is sent over LoRaWAN to all repellers that are assigned to that specific camera.
  4. After activating the repeller(s), the system sends an alert to the forest rangers, notifying them of a bear detection.
A repeller

My contributions to this project were:

There were many diverse challenges to overcome in this project, ranging from software to hardware. This great variety of challenges made it a very enjoyable project to work on.

Placing a cameratrap on a treePlacing a cameratrap on a treeA repeller in a treeBattery casesDeployment of the repellersDeployment of the repellersDeployment of the repellers