AI Scarecrow

An adaptive AI scarecrow that detects predators and intruders and dynamically scares them away to protect livestock.

Why This Matters

This project was built for Hack Canada 2026 under the theme “Solving Problems Canadians Face.”

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.

Our scarecrow explores a low-cost, AI-powered deterrent system that adapts to different threats in real time while remaining humane to wildlife.

Instead of static scare tactics, our scarecrow detects what is approaching and deploys the most effective deterrent for that specific animal or intruder.

What It Does

Our scarecrow is a smart farm sentinel that:

• Detects animals and humans using a webcam and ML detection
• Identifies species (e.g. raccoon, coyote, birds, humans)
• Deploys adaptive deterrents such as predator sounds or voice warnings for human intruders
• Logs detections and responses in a live monitoring dashboard
• Uses AI voice generation for human intruder warnings

Example Deterrents

• Crow → Hawk screech
• Raccoon → Dog bark
• Deer → Human voice warning
• Human intruder → Custom human voice warning

Dashboard Features

• Live webcam feed
• Detection log with timestamps and deterrent action
• Manual scare soundboard

How to Run

Backend

1. Create a .env file in the backend/ directory with your API keys:

   GEMINI_API_KEY=key
   ELEVENLABS_API_KEY=key
   

2. Navigate to the backend directory and sync dependencies:

   cd backend
   uv sync

3. Start the FastAPI server:

   fastapi dev

4. In a new terminal, activate the virtual environment and run the ML detection:

   cd backend/ml
   source venv/bin/activate
   python yolo.py --camera 0  # or --camera 1 depending on your webcam

Frontend

1. Navigate to the frontend directory and install dependencies:

   cd frontend
   pnpm install

2. Start the development server:

   pnpm run dev

3. Open your browser to the URL shown in the terminal (usually http://localhost:5173)

Why Adaptive Scarecrows?

Traditional scarecrows fail because animals quickly become accustomed to them.

Our scarecrow adapts by:

• Detecting the specific animal species
• Selecting the most effective deterrent
• Rotating deterrents to prevent habituation
• Logging encounters to understand farm threat patterns

Why This Is Relevant in Canada

• Canada has massive rural farmland regions
• Farms are often located near forests and wildlife habitats
• Livestock losses from predators are a recurring issue
• Humane wildlife deterrence is preferred over harmful control methods

Common farm predators include:

• Coyotes
• Foxes
• Raccoons
• Birds
• Deer

Tech Stack

Frontend:

• React
• Vite
• pnpm
• WebSockets

Backend:

• ML object detection
• Webcam video input
• Event streaming to frontend

Integrations:

• Gemini
• ElevenLabs voice generation

Hardware (Hackathon MVP)

• Tripod scarecrow stand
• Halloween mask
• Bluetooth speaker
• Logitech Brio webcam

Demo

1. Webcam detects animal or human
2. ML identifies the species
3. Our scarecrow selects a deterrent
4. Sound or voice plays through the scarecrow
5. Event appears on the dashboard

Future Ideas

• Farm-wide threat heatmaps
• Solar-powered remote units
• Multiple scarecrow network across fields
• Mobile alerts for farmers
• Learning which deterrents work best

---

Built at Hack Canada 2026 in Waterloo 🇨🇦