🤖 FaceTracker: Standalone ESP32-CAM System

A self-contained computer vision solution that tracks human faces in real-time. This project transforms the ESP32-CAM into an active robotic sensor that independently calculates coordinates—completely independent of external apps.

No Middleman: This system processes detection onboard and acts immediately, functioning without any Wi-Fi connection.

Get Source Code

Full Build Instructions & ESP-WHO Logic

Hardware Configuration

How It Works: The Logic Flow

The system operates in a continuous feedback loop to maintain target alignment:

• Capture: The OV2640 sensor grabs a frame at QVGA resolution (320 x 240 pixels).
• Detection: The system scans for face landmarks internally using the ESP-WHO framework.
• Displacement Calculation: It determines the centroid coordinate distance from the frame center point (160, 120).
• Actuation: PWM adjustment commands are sent via the 400kHz I2C bus to the PCA9685 driver to re-center the camera.

Technical Control Loop & Logic

The software calculates the positional offset required to drive the Pan/Tilt motors with high precision:

• Pan Offset: Calculated by subtracting the target's current X-coordinate from the frame center.
• Tilt Offset: Calculated by subtracting the target's current Y-coordinate from the frame center.
• Hardware Acceleration: Offloading 12-bit PWM processing to the PCA9685 prevents CPU bottlenecks and eliminates servo jitter common in software-timed pulses.

Technical Specifications

🗺️ Future Roadmap

• S3-Series Migration: Moving to the ESP32-S3 CAM for higher FPS, while tackling new pinout challenges.
• PID Control: Implementing smoother derivative logic for fluid tracking.
• Target Prioritization: Smart logic to follow the largest subject in crowded frames.