Published December 15, 2023 © GPL3+

AI Face Recognition with Jetson Nano

This is an example of using a Jetson Nano with OpenCV and Python's face-recognition library to train a model and recognize faces in photos.

IntermediateFull instructions provided4 hours999

Things used in this project

Hardware components

NVIDIA Jetson Nano Developer Kit

Software apps and online services

OpenCV – Open Source Computer Vision Library OpenCV

Story

Project Goals

Build a Python program that can recognize faces in photos
Train the program on known faces
Recognize and identify those trained faces in random photos

Tech Stack

Ubuntu 18.04 LTS
Python 3
Nvidia Jetpack 4.3-b134
OpenCV open source library for computer vision 4.1.1
Python open source face-recognition library
Dlib machine learning toolkit
Pickle Python object serializer

Setup

Installed Nvidia Jetpack on and SD card and installed in the Jetson Nano
Installed OpenCV
Installed Dlib
Installed face-recognition library
Installed pickle

Training and Testing Images

I started by downloading training images of my two subjects, Bill Gates and Steve Jobs and placed them in a "known" images folder. Then, I downloaded several other images with Bill and/or Steve and other people and placed those in an "unknown" folder.

Program Workflow

Using the Python face-recognition library, the program loads each training image and encodes it using the file name as the label. Then, using pickle, the training data is saved into a file named train.pkl.

Next, the face-recognition library is used to decode each test image, locate faces in the images, and encode them. Then, the encoded testing images are compared to the training images and rectangles and labels are applied using OpenCV. Finally, the new images with annotations are saved.

Results

Conclusions

The results were not perfect, but considering I only trained the model on one image of each person, it did a pretty good job! I was surprised it recognized their faces when they were much younger and picked them out with different expressions and backgrounds.

AI Face Recognition with Jetson Nano

# AI Face Recognition with Jetson Nano
# Using OpenCV and face_recognition to identify known faces
# Mike Soniat

import face_recognition
import cv2
import os
import pickle 

Encodings=[]
Names=[]

# comment out training code after saving with Pickle
# train on known images
print("Training...")
known_dir = '/home/mikes/Desktop/PyPro/faceRecognizer/demoImages/known'
for root, dirs, files in os.walk(known_dir):
    for file in files:        
        path = os.path.join(root, file)
        name = os.path.splitext(file)[0]
        print(name)
        person = face_recognition.load_image_file(path)
        encoding = face_recognition.face_encodings(person)[0]
        
        # load arrays
        Encodings.append(encoding)
        Names.append(name)

# save training data using pickle
with open('train.pkl','wb') as f:
    pickle.dump(Names, f)
    pickle.dump(Encodings, f)

# read training data from pickled file
with open('train.pkl','rb') as f:
    Names = pickle.load(f)
    Encodings = pickle.load(f)

# set font for name labels
font=cv2.FONT_HERSHEY_SIMPLEX

# step through unknown files
unknown_dir = '/home/mikes/Desktop/PyPro/faceRecognizer/demoImages/unknown'
for root, dirs, files in os.walk(unknown_dir):
    for file in sorted(files):
        print(file)
        testImagePath = os.path.join(root, file)
        testImage = face_recognition.load_image_file(testImagePath)

        # find faces in unknown pic and encode
        face_positions = face_recognition.face_locations(testImage)
        allEncodings = face_recognition.face_encodings(testImage, face_positions)

        # convert pic to BGR (for OpenCV)
        testImage = cv2.cvtColor(testImage, cv2.COLOR_RGB2BGR)

        # step through faces and names in unknown pic
        for (top,right,bottom,left), face_encoding in zip(face_positions, allEncodings):
            name = 'Unknown Person'
            matches = face_recognition.compare_faces(Encodings, face_encoding)
            if True in matches:
                first_match_index = matches.index(True)
                name = Names[first_match_index]

            cv2.rectangle(testImage, (left,top),(right,bottom),(0,0,255), 2)
            cv2.putText(testImage, name, (left,top-6), font, 1, (0,255,255), 2)

        # show next image in window
        cv2.imshow(file, testImage)
        cv2.moveWindow(file,0,0)
        
        # wait for user to press a key
        cv2.waitKey(0)
        cv2.destroyAllWindows()