Face Detection & Recognition using Yolo, ResNet-18 & Arduino

A Python-based face detection system using OpenCV and YOLOv3 for face detection, integrated with Arduino for real-time signal transmission, and a Python-based face recognition system leveraging a pre-trained ResNet-18, also integrated with Arduino for signal output.

Arduino Setup

Arduino Circuit Setup

Arduino Code in C Language

bool status = false;

void setup() {
    pinMode(13, OUTPUT); 
    pinMode(12, OUTPUT);
    Serial.begin(9600);
}

void loop() {
    if (Serial.available() > 0) {
        char received = Serial.read();
        if (received == '1') {
            status = true;
        } else if (received == '0') {
            status = false;
        }
    }
    
    digitalWrite(12, status ? HIGH : LOW);
    digitalWrite(13, status ? LOW : HIGH);
}

Arduino Circuit in Action when data is sent using the open stream

YOLO Face Detection

This project uses YOLOv3-Face for real-time face detection, which was trained on the WIDER face dataset.

1. Download Model Weights

Click here to download the model weights

2. Install Dependencies

pip install opencv-python numpy

3. Load YOLO & Webcam

import cv2
import numpy as np

net = cv2.dnn.readNet("yolov3-wider_16000.weights", "yolov3-face.cfg")
cap = cv2.VideoCapture(0)

4. Optimized Face Detection

while cap.isOpened():
    ret, frame = cap.read()
    if not ret:
        break

    blob = cv2.dnn.blobFromImage(frame, 1/255.0, (416, 416), swapRB=True, crop=False)
    net.setInput(blob)
    outputs = net.forward(net.getUnconnectedOutLayersNames())

    boxes, confidences = [], []
    for output in outputs:
        for detection in output:
            confidence = max(detection[5:])
            if confidence > 0.5:
                x, y, w, h = (detection[:4] * [frame.shape[1], frame.shape[0], frame.shape[1], frame.shape[0]]).astype(int)
                boxes.append([x, y, w, h])
                confidences.append(float(confidence))

    for i in cv2.dnn.NMSBoxes(boxes, confidences, 0.5, 0.4).flatten():
        x, y, w, h = boxes[i]
        cv2.rectangle(frame, (x, y), (x + w, y + h), (0, 255, 0), 2)

    cv2.imshow("YOLO Face Detection", frame)
    if cv2.waitKey(1) & 0xFF == ord('q'):
        break

cap.release()
cv2.destroyAllWindows()

Face Recognition

This project implements a real-time face recognition system using OpenCV, PyTorch, and a ResNet-18 model as a feature extractor. It consists of two scripts:

1. register.py – Captures a person's face, extracts embeddings using ResNet-18, and saves them.
2. recognize.py – Detects faces in a video stream, compares embeddings with saved ones, and identifies the person.

How It Works (Concise Version):

1. Face Detection – Uses OpenCV’s Haarcascade to detect faces from the webcam feed.
2. Feature Extraction – A pretrained ResNet-18 (with the last layer removed) converts faces into embeddings.
3. Face Registration (register.py) – Captures a face, extracts its embedding, and saves it as a .npy file.
4. Face Recognition (recognize.py) – Compares the detected face’s embedding with saved ones using cosine similarity.
   • If similarity is below 0.1, the person is recognized; otherwise, labeled "Unknown".
   • Sends a signal to an Arduino (1 for recognized, 0 for unknown).

This system enables real-time face detection using YOLOv3 & face recognition with pre-trained ResNet-18, triggering the Arduino's LED when a face is detected. The integration ensures minimal lag and efficient performance, making it suitable for various applications.