ESP32-CAM AI Object Tracking Camera

This project is a prototype AI object-tracking camera built with an ESP32-CAM, a second ESP32, a PCA9685 servo driver, two MG995/MG996-style servos, a custom YOLO model, and 3D-printed mechanical parts.

The ESP32-CAM streams live video to my laptop over Wi-Fi. Python runs a trained YOLO model to detect my custom 3D-printed target object. Then Python sends movement commands to a second ESP32, which controls the pan and tilt servos through the PCA9685 servo driver.

The goal of this project was to connect AI object detection with real hardware movement and design my own mechanical parts instead of using a ready-made pan-tilt mount.

Demo

Project photo:

How It Works

ESP32-CAM
   ↓ Wi-Fi video stream
Laptop running Python + YOLO
   ↓ serial movement command
Second ESP32
   ↓ I2C
PCA9685 servo driver
   ↓
Pan and tilt servos
   ↓
Camera follows the object

Hardware

ESP32-CAM
ESP32-WROOM-32 DevKit board
PCA9685 16-channel servo driver
2 MG995/MG996-style servos
External 5V power supply for the servos
Jumper wires
M3 screws
Micro screws for the ESP32-CAM module
3D-printed camera mount parts

Wiring

Basic wiring:

ESP32 GPIO21 → PCA9685 SDA
ESP32 GPIO22 → PCA9685 SCL
ESP32 5V → PCA9685 VCC
ESP32 GND → PCA9685 GND
External 5V → PCA9685 V+
External power GND → PCA9685 GND
Pan servo → PCA9685 channel 0
Tilt servo → PCA9685 channel 1

The PCA9685 uses I2C communication. SDA is the data line, and SCL is the clock line.

The servos are powered with external 5V power because servos need more current than the ESP32 should provide directly.

See wiring diagram:

3D Printed Parts

The STL files are in the stl/ folder.

The OpenSCAD source file is in the openscad/ folder, so the design can be adjusted for different hardware.

This design was made for my specific prototype, so dimensions may need to be adjusted depending on your servo, camera board, screws, and printer tolerances.

AI Model

I trained a custom YOLO model using my own images of a 3D-printed target object.

The trained model file is:

models/best.pt

Python Setup

Install the Python dependencies:

pip install -r requirements.txt

Update these values in python/ai_tracking.py for your own setup:

MODEL_PATH = "../models/best.pt"
SERIAL_PORT = "YOUR_SERIAL_PORT"
CAMERA_URL = "http://YOUR_ESP32_CAM_IP:81/stream"

Run the main tracking script:

cd python
python3 ai_tracking.py

Arduino Setup

Upload the ESP32-CAM streaming code to the ESP32-CAM:

arduino/esp32_cam_stream/esp32_cam_stream.ino

Upload the servo controller code to the second ESP32:

arduino/servo_controller/servo_controller.ino

Before uploading the ESP32-CAM code, replace the Wi-Fi placeholders with your own Wi-Fi name and password.

Project Files

python/
  capture_images.py
  prepare_dataset.py
  ai_tracking.py

arduino/
  esp32_cam_stream/
  servo_controller/

stl/
  3D-printable parts

openscad/
  editable OpenSCAD source

docs/
  wiring diagram and notes

models/
  trained YOLO model

Testing

I also included a testing checklist for the full prototype behavior, including startup, tracking, lost-target behavior, servo movement, latency, edge limits, long-run stability, and thermal checks.

See: Testing Checklist

Notes

This is still a prototype. The movement, wiring, 3D-printed parts, and code are still being improved.

For real-time tracking, a smaller and faster video stream worked better than a high-resolution stream with too much delay.

License

This project is shared for learning and experimentation.

ESP32-CAM-AI-Object-Tracking

README