AI Vision Machine, a ESP32 project

This project builds a face recognition system using the ESP32 microcontroller and DFRobot HUSKYLENS 2 AI Vision Sensor. The HUSKYLENS handles all the computer vision processing on its own, making it easy to add machine learning capabilities without complex algorithms or training pipelines.

The guide provides a complete parts list, wiring diagram showing power and I2C signal connections, and ready-to-use firmware that recognizes faces and reports their position on screen. After assembly, teach the camera a face by following the on-device prompts, then point it at that person to see real-time recognition results streamed over serial.

Wiring diagram

Gather all the parts

Qty	Component
1	DFRobot HUSKYLENS 2 AI Vision Sensor HUSKYLENS 2 A small smart camera that recognizes faces locally and sends the results to the ESP32.

Assemble it in 4 steps

1. Place the two modules

Put the ESP32 DevKit v1 and the HUSKYLENS 2 where you can reach both sets of header pins. Leave the ESP32 unplugged while you add wires, so a loose wire cannot accidentally touch the wrong pin.

Keep the labels on both boards facing upward so you can read VCC, GND, SDA, and SCL.
Do not power either board while moving jumper wires; a misplaced power wire can damage a module.

2. Connect power and ground

Use one red jumper wire from the HUSKYLENS 2 VCC pin to the ESP32 3V3 pin (power). Use one black jumper wire from HUSKYLENS 2 GND to an ESP32 GND pin (ground). These wires give the camera safe 3.3-volt power and a shared return path.

Make sure VCC and GND are not swapped — the printed labels, not the header position, are what matter.
Swapped power wires can permanently damage the HUSKYLENS camera.

3. Connect the two signal wires

Connect HUSKYLENS 2 SDA to ESP32 GPIO21 (data), then connect HUSKYLENS 2 SCL to ESP32 GPIO22 (clock). These two wires let the camera send recognition results to the ESP32.

Use two different wire colors for SDA and SCL so a crossed pair is easy to spot later.
Do not connect the HUSKYLENS signal pins to 5V; the ESP32 uses 3.3-volt signals.

4. Teach a face to the camera

After the firmware is deployed, use the HUSKYLENS 2 screen controls to select Face Recognition and learn a face. Hold the person’s face clearly in the camera view while teaching it. The learned number is the ID that the ESP32 prints when it recognizes that person.

Use even lighting and keep the camera steady while teaching for more reliable recognition.
Only enroll people who have agreed to be recognized, and do not use the device for security-critical identity decisions.

Review all connections

1. Connections between "huskylens_1" and "ESP32"

Function	huskylens_1	ESP32
`power`	VCC	3V3
`ground`	GND	GND
`i2c`	SDA	GPIO 21
`i2c`	SCL	GPIO 22

Deploy the firmware

main.cppOpen in Schematik

#include <Arduino.h>
#include <Wire.h>
#include <DFRobot_HuskylensV2.h>


// Forward declarations
void printIfChanged(const String &message);

constexpr uint8_t HUSKYLENS_SDA = 21;
constexpr uint8_t HUSKYLENS_SCL = 22;
constexpr uint32_t REPORT_INTERVAL_MS = 250;

DFRobot_HuskylensV2 huskylens;
String lastMessage;
uint32_t lastReportMs = 0;

void printIfChanged(const String &message) {
  if (message != lastMessage) {
    Serial.println(message);
    lastMessage = message;
  }
}

void setup() {
  Serial.begin(115200);
  Wire.begin(HUSKYLENS_SDA, HUSKYLENS_SCL);

  Serial.println("AI camera starting...");
  while (!huskylens.begin(Wire)) {
    Serial.println("HUSKYLENS not found. Check VCC, GND, SDA, and SCL.");
    delay(1000);
  }

  huskylens.switchAlgorithm(ALGORITHM_FACE_RECOGNITION);
  Serial.println("Ready. Teach a face on the HUSKYLENS screen, then point it at that person.");
}

void loop() {
  if (millis() - lastReportMs < REPORT_INTERVAL_MS) {
    return;
  }
  lastReportMs = millis();

  if (huskylens.getResult(ALGORITHM_FACE_RECOGNITION) != 0) {
    printIfChanged("Camera is connected, but no face result is available.");
    return;
  }

  FaceResult result = huskylens.readFaceResult();
  if (result.ID > 0) {
    String message = "Recognized face ID " + String(result.ID) +
                     " at x=" + String(result.xCenter) +
                     ", y=" + String(result.yCenter) +
                     ", size=" + String(result.width) + "x" + String(result.height);
    printIfChanged(message);
  } else {
    printIfChanged("Face detected, but it has not been taught a name or ID yet.");
  }
}