Real‑Time Face Recognition on Raspberry Pi – End‑to‑End Guide

This tutorial walks you through building a real‑time face‑recognition system on a Raspberry Pi using OpenCV and a Pi‑Cam. From hardware setup to training the recognizer, every step is presented with clear code snippets and practical tips.

Below you’ll see a live demo of the Pi‑Cam detecting faces in real time.

The project relies on the open‑source OpenCV library, which is optimized for real‑time performance. While we’ll focus on a Raspberry Pi running Raspbian, the same code runs on macOS and Windows with only minor adjustments.

3 Phases

A complete face‑recognition workflow can be broken into three distinct stages:

• Face Detection & Data Gathering
• Training the Recognizer
• Real‑time Face Recognition

The diagram below summarizes these phases.

Step 1: Bill of Materials

Essential components:

• Raspberry Pi 3 – $32.00
• 5‑MPx 1080p OV5647 Mini Camera Module – $13.00

Step 2: Installing OpenCV 3

We recommend following Adrian Rosebrock’s guide for a clean OpenCV 3 installation on Raspbian Stretch. After the tutorial, activate the dedicated Python virtual environment:

source ~/.profile
workon cv

Once inside the environment, verify the Python version and OpenCV installation:

python
>>> import cv2
>>> cv2.__version__

You should see a version number like 3.3.0 or newer.

Step 3: Testing Your Camera

If the camera is not detected, enable the driver with:

sudo modprobe bcm2835-v4l2

The following Python snippet opens the Pi‑Cam stream, displays it in color and grayscale, and flips it vertically (adjust if your camera orientation differs):

import numpy as np
import cv2
cap = cv2.VideoCapture(0)
cap.set(3, 640)
cap.set(4, 480)
while True:
    ret, frame = cap.read()
    frame = cv2.flip(frame, -1)
    gray = cv2.cvtColor(frame, cv2.COLOR_BGR2GRAY)
    cv2.imshow('frame', frame)
    cv2.imshow('gray', gray)
    if cv2.waitKey(30) & 0xff == 27:
        break
cap.release()
cv2.destroyAllWindows()

Run it with python simpleCamTest.py (available on the project’s GitHub). Press ESC to exit.

Step 4: Face Detection

OpenCV’s Haar Cascade classifier is a lightweight, machine‑learning approach for detecting faces. Download the pre‑trained XML file (haarcascade_frontalface_default.xml) and place it in a Cascades folder. The following script demonstrates real‑time face detection with bounding boxes:

import cv2
faceCascade = cv2.CascadeClassifier('Cascades/haarcascade_frontalface_default.xml')
cap = cv2.VideoCapture(0)
cap.set(3, 640)
cap.set(4, 480)
while True:
    ret, img = cap.read()
    img = cv2.flip(img, -1)
    gray = cv2.cvtColor(img, cv2.COLOR_BGR2GRAY)
    faces = faceCascade.detectMultiScale(
        gray,
        scaleFactor=1.2,
        minNeighbors=5,
        minSize=(20, 20)
    )
    for (x, y, w, h) in faces:
        cv2.rectangle(img, (x, y), (x + w, y + h), (255, 0, 0), 2)
    cv2.imshow('video', img)
    if cv2.waitKey(30) & 0xff == 27:
        break
cap.release()
cv2.destroyAllWindows()

The detectMultiScale parameters control sensitivity: scaleFactor reduces image size each iteration, minNeighbors filters false positives, and minSize sets the smallest detectable face.

Additional Examples

Explore more detection options on the project’s GitHub:

• faceEyeDetection.py
• faceSmileDetection.py
• faceSmileEyeDetection.py

Haar Cascade Object Detection – Face & Eye Tutorial

Step 5: Data Gathering

Collecting a labeled dataset is essential for training. Create a project folder (e.g., FacialRecognitionProject) and a dataset subdirectory:

mkdir FacialRecognitionProject
cd FacialRecognitionProject
mkdir dataset

Download the Haar Cascade file into the project root. Then run 01_face_dataset.py to capture 30 face images per user:

import cv2
import os
cam = cv2.VideoCapture(0)
cam.set(3, 640)
cam.set(4, 480)
face_detector = cv2.CascadeClassifier('haarcascade_frontalface_default.xml')
face_id = input('\nEnter user ID (numeric) and press <Return> → ')
print('\n[INFO] Initializing face capture. Look at the camera…')
count = 0
while True:
    ret, img = cam.read()
    img = cv2.flip(img, -1)
    gray = cv2.cvtColor(img, cv2.COLOR_BGR2GRAY)
    faces = face_detector.detectMultiScale(gray, 1.3, 5)
    for (x, y, w, h) in faces:
        cv2.rectangle(img, (x, y), (x + w, y + h), (255, 0, 0), 2)
        count += 1
        cv2.imwrite(f'dataset/User.{face_id}.{count}.jpg', gray[y:y+h, x:x+w])
        cv2.imshow('image', img)
        if cv2.waitKey(100) & 0xff == 27 or count >= 30:
            break
    if count >= 30:
        break
cam.release()
cv2.destroyAllWindows()
print('\n[INFO] Exiting Program and cleaning up.')

Each image is named User.{id}.{count}.jpg (e.g., User.1.4.jpg). Adjust the sample count as needed.

Step 6: Trainer

Using the collected images, train an OpenCV face recognizer. The resulting model is saved as a .yml file in a trainer folder. Refer to the full training script in the project repository for implementation details.

Read More Detail: Real‑Time Face Recognition – End‑to‑End Project