🤖 FaceTracker: Standalone ESP32-CAM Pan-Tilt System

FaceTracker is an autonomous Edge Computing solution for real-time computer vision. By integrating an ESP32-CAM with a PCA9685 PWM controller, the system performs onboard facial centroid tracking—translating visual coordinates directly into physical motion without the need for external processing or Wi-Fi.

---

🚀 The Backstory: Edge AI

Most ESP32-CAM implementations are limited to "CameraWebServer" streams. This project moves the intelligence to the Edge, allowing the microcontroller to scan, detect, and act autonomously. By eliminating the latency of a web browser middleman, the system calculates positional error vectors in real-time. The result is a self-contained robotic sensor that centers on human subjects using high-speed I2C communication, making it ideal for mobile robotics and interactive animatronics.

---

✨ Key Technical Features

• Autonomous Centroid Tracking: Onboard processing calculates the X/Y displacement required to maintain target alignment at the frame center.
• Low-Latency Edge AI: Uses the MTMN (Multi-Task Cascaded Convolutional Networks) model for face detection without cloud or app dependency.
• Hardware-Accelerated PWM: Offloads servo pulses to a PCA9685 via 400kHz Fast-Mode I2C, ensuring 12-bit resolution and jitter-free movement.
• Real-Time Telemetry: Local SSD1306 OLED integration provides instant feedback on coordinate mapping and system health.
• Optimized Bus Architecture: Maximizes the ESP32-CAM's limited GPIO by daisy-chaining all peripherals on a single I2C bus.

🛠️ Hardware Stack

• ESP32-CAM (AI-Thinker):
  • Dual-core LX6 CPU @ 240MHz.
  • Integrated OV2640 image sensor.
• PCA9685 16-Channel PWM Driver:
  • 12-bit resolution for precise angular servo increments.
  • Addresses I2C bus bottlenecking by managing pulse timing independently.
• Actuators: Dual SG90/MG90S Micro Servos for Pan and Tilt.
• Display: SSD1306 OLED for coordinate debugging.

---

🔌 Wiring & Pinout (I2C Configuration)

All external peripherals are synchronized via the I2C Serial Bus to conserve GPIO pins for the camera interface.

Peripheral	ESP32-CAM Pin	Function	Protocol
PCA9685	GPIO 14	SDA	I2C Data
PCA9685	GPIO 15	SCL	I2C Clock
SSD1306 OLED	GPIO 14	SDA	Shared Bus
SSD1306 OLED	GPIO 15	SCL	Shared Bus
Logic Power	5V / GND	VCC / GND	Logic Supply

Important

Servo Power: Always use a dedicated 5V power supply for the PCA9685/Servos. Drawing servo current through the ESP32-CAM will cause brownouts and core resets.

---

📐 Logic & Control Loop

Centroid Mapping

The system identifies facial landmarks and determines the Error Vector relative to the QVGA resolution center ($160, 120$).

• The Goal: Minimize the delta between the detected face centroid and the frame origin.
• The Action: The ESP32 calculates a proportional correction and updates the PCA9685 registers to adjust the servo duty cycle.

Data Pipeline

1. Acquisition: Frame capture via OV2640 DMA.
2. Inference: MTMN model detects faces and produces bounding box coordinates.
3. Control: Software calculates $X, Y$ offsets and generates I2C commands.
4. Feedback: Coordinates are pushed to the OLED and Serial Monitor.
5. Actuation: Servos pivot to center the target within the frame.

---

⚡ Quick Start

1. Hardware: Assemble the circuit using the shared I2C pinout.
2. Project Initialization:
   • Create a new project in PlatformIO.
   • Select AI Thinker ESP32-CAM as the board.
3. Critical Configuration Steps: This project uses a legacy framework (Espressif v1.0.4) that requires manual file handling to resolve missing fd_forward.h and ADS1115-Driver errors. Complete these three steps before attempting to compile:
   • 1. Replace platformio.ini: Overwrite the default file in your project root with the specific configuration provided (ensuring it uses platform = espressif32@1.12.4).
   • 2. Replace main.cpp: Copy your code into the src folder, replacing the existing boilerplate file.
   • 3. Manual Library Install: Copy the ADS1115_Driver-1.0.2 and esp32cam folders into the project’s lib folder. This allows the missing legacy files to be available for the compiler.
4. Calibration: Define SERVO_MIN and SERVO_MAX within the code for physical calibration.
5. Build & Upload: Connect your ESP32-CAM via an FTDI adapter and click Upload in PlatformIO.

---

📂 Required Folder Structure

To ensure the compiler sees your files, your project directory should look like this:

Your_Project_Folder/
├── lib/
│   ├── ADS1115_Driver-1.0.2/
│   └── esp32cam/            <-- Contains legacy fd_forward.h
├── src/
│   └── main.cpp             <-- Your actual code
└── platformio.ini           <-- The specific v1.12.4 config

---

🔍 Troubleshooting

• I2C Communication: If the OLED or PCA9685 is not detected, verify that GPIO 14/15 are not being pulled high/low by other peripherals.
• FPS Drop: Ensure "Core Debug Level" is set to "None" in the IDE to maximize CPU cycles for detection logic.
• Servo Jitter: Confirm the ground (GND) is common between the ESP32-CAM and the external servo power supply.

🗺️ Future Roadmap

• ESP32-S3 Port: Leveraging vector instructions for higher frame rates.
• PID Implementation: Replacing simple proportional tracking with a full Proportional-Integral-Derivative loop for fluid motion.
• Distance Logic: Using bounding box area to estimate target distance and adjust tracking speed dynamically.

---

© 2026 MatsRobot | Licensed under the MIT License