Teachable Machine + Wokwi Serial Control

Created by Yogesh M

---

🚀 Overview

This project integrates a Teachable Machine Image Model with the Web Serial API to send real-time predictions to a Wokwi simulation (or any microcontroller via serial).

• Uses a webcam to capture live images
• Runs an image classification model in the browser
• Detects the most probable class
• Sends the detected class to Wokwi via serial

---

🧠 Features

• Real-time image classification
• Confidence-based filtering (only sends if probability > 0.70)
• Serial communication using Web Serial API
• Simple browser-based interface

---

📁 Project Structure

project-folder/
│
├── index.html
├── README.md
└── my_model/
    ├── model.json
    ├── metadata.json
    └── weights.bin

---

⚙️ Requirements

• Google Chrome (Web Serial API supported)
• Teachable Machine exported image model
• Wokwi simulation OR physical device with serial support

---

🔌 How It Works

1. Connect to Wokwi

• Click Connect Wokwi
• Select the serial port
• Connection opens at 115200 baud

2. Start Model

• Click Start Model
• Loads the trained model
• Starts webcam
• Begins prediction loop

3. Prediction Logic

• Evaluates all classes
• Picks highest probability
• If probability > 0.70 → sends class name via serial

---

🖥️ Usage

Run the Project

Open index.html in Chrome

If opening locally, use a local server (e.g., Live Server)

---

Start Workflow

1. Connect Wokwi
2. Start Model
3. Show objects to webcam
4. Observe predictions + serial output

---

📡 Serial Output Format

ClassName\n

Example:

LED_ON
LED_OFF

---

🔧 Customization

Change confidence threshold

if (writer && bestProb > 0.70)

Modify serial message

bestClass + "\n"

---

⚠️ Notes

• Works only on https or localhost
• Camera permission required
• Ensure correct path for my_model/

---

💡 Use Cases

• Gesture-based control
• Smart IoT automation
• Object-triggered actions
• AI-powered device control

---

📌 Future Improvements

• Add UI indicators for detected class
• Map classes to specific hardware actions
• Add audio/visual feedback
• Improve model accuracy