Scarecrow

GIF
Lit up scarecrow on 4th floor

Inspiration

Canada has vast rural farmland that sits alongside dense wildlife populations. Farmers regularly lose livestock and crops to predators and pests, and traditional scarecrows quickly become ineffective as animals learn to ignore them.

Tens of thousands of Canadian livestock die each year. In Ontario alone, 2,537 livestock were killed by wildlife predators in 2024, mostly by coyotes
Canadian farms have many different predators like: coyotes, wolves, bears, foxes, raccoons, ravens, etc. meaning deterrence tech should be adaptive
Canada prefers humane wildlife management, and our government funds programs to prevent predation rather than to kill predators - this makes deterrence tech more valuable than trapping/poisoning animals

We also arrived late (on Saturday night) so if we wanted to compete against 800 hackers, we needed something unconventional.

Thus, a scarecrow was constructed in the 4th floor sleeping area at night.

What it does

Our AI Scarecrow is an adaptive farm surveillance system that detects predators and intruders and dynamically scares them away.

Using a camera and ML detection, our scarecrow identifies animals or humans approaching livestock and deploys the most effective deterrent based on the target.

How we built it

Hardware (very scientific):

Tripod
Halloween mask (facebook marketplace)
Bluetooth speaker
Logitech Brio webcam

Frontend

React + Vite, PNPM, Tailwind for the dashboard
WebSocket event stream
Live video feed and detection logs

Backend

ML animal/human detection on webcam frames
Animal species classification
Event streaming to the frontend

Integrations

Gemini for reasoning and threat classification
ElevenLabs for generating human voice deterrents

When an animal or intruder is detected:

The ML model identifies the species
The backend chooses an appropriate deterrent
The scarecrow plays the sound through the speaker
The event is logged on the dashboard

Challenges we ran into

Building a physical scarecrow
Getting ML detection running reliably on live webcam footage
Making people too afraid to sleep in the dedicated sleeping area
Adding cooldown logic to not exhaust the backend

Accomplishments that we're proud of

Protecting the livestock and crops of the Waterloo region
Finishing a working demo after starting the project a day late
Intimidating hackers with our 6'2 scarecrow

What we learned

Hardware + software demos are extremely fun
Real-time monitoring systems require putting thought into optimization (especially when using APIs)
We need to start fundraising for this immediately