ESP32-CAM Face Detection using TinyML + Flask Dashboard

Overview

This project demonstrates a real-time face detection system using:

• ESP32-CAM (without PSRAM) for capturing images
• TinyML model (.tflite) running on a Python Flask backend
• A web dashboard for displaying real-time detection and history
• Telegram bot integration for sending face detection alerts with image, timestamp, and location

Designed for low-resource hardware, this solution is lightweight and ideal for entry-level IoT AI applications.

---

Components Used

• ESP32-CAM (AI Thinker)
• TensorFlow Lite (TFLite) for model inference
• Flask Web Framework (Python)
• OpenCV for image handling
• Telegram Bot API for notifications

---

---

Features

• Real-time image capture from ESP32-CAM
• Face detection using quantized TFLite model
• Web dashboard to view latest image + status
• Instant Telegram alerts with location + timestamp
• Training notebook for building your own model
• Fast, works with low-resolution images (160x120)

---

Setup Guide

1. Flash ESP32-CAM

• Open esp32/esp32_cam_sender.ino in Arduino IDE
• Update WiFi credentials and Flask server IP:

const char* ssid = "YOUR_WIFI_NAME";
const char* password = "YOUR_PASSWORD";
const char* serverUrl = "http://<your-pc-ip>:8000/upload";

• Set board to AI Thinker ESP32-CAM and flash the code

2. Set Up Python Server

cd flask_app pip install -r requirements.txt python app.py

Make sure face_detection.tflite is in flask_app/model/

Visit http://localhost:8000 to open the dashboard.

---

Telegram Alerts

How to Enable

1. Create a bot via @BotFather
2. Get your bot token and your own chat ID
3. Replace in app.py:

TELEGRAM_BOT_TOKEN = 'your_bot_token' TELEGRAM_CHAT_ID = 'your_chat_id' LOCATION = "Lab Entrance" # Customizable tag

---

Train Your Own Face Detection Model

• Open notebook/train_and_convert.ipynb
• Use a dataset like WIDER FACE or your custom face images
• Build a lightweight CNN model
• Export and quantize to TensorFlow Lite:

converter = tf.lite.TFLiteConverter.from_keras_model(model) converter.optimizations = [tf.lite.Optimize.DEFAULT] tflite_model = converter.convert()

• Save as face_detection.tflite and move it to flask_app/model/

---

Web Dashboard

Dashboard (/) shows:

• Latest image with face bounding boxes
• Detection status (face/no-face)
• Last detected time
• Auto-refreshing UI

---

Testing

1. Power up ESP32-CAM
2. Server receives POST image every ~100ms
3. Flask runs inference on image using TFLite
4. If face is found:
   • Detection shown on dashboard
   • Telegram alert sent with image

---

Requirements :

Flask==2.3.2
opencv-python==4.8.0.74
numpy==1.24.3
requests==2.31.0
tensorflow==2.12.0