Real-time facial recognition with python dlib v2.py

# -*- coding: utf-8 -*-
"""
The following python script utilizes the dlib library for facial recognition.
 The script first reads in a database of faces stored in a saved CSV file. 
 
 The database contains face descriptors and bounding boxes for each face detected in a set of images.
Once the database has been loaded, the script then uses the dlib library to 
perform real-time facial recognition on a live stream from a PC webcam. 

The webcam captures images in real-time, and the dlib library is used to detect faces in each image 
and compare them to the faces in the database.

The script implements a brute nearest neighbor search algorithm to compare 
the faces detected in the webcam stream to the faces in the database. 

The algorithm calculates the distance between the face descriptors for each pair of faces
 and returns the closest match. 
 
 If a match is found, the script will display the name of the person in the database 
 that the live face is closest to.
 
This script provides a demonstration of how the dlib library can be used for real-time facial recognition. 
The use of a saved CSV file for storing the face database allows for a flexible and scalable solution, 
as the database can be easily updated or expanded as needed. 
"""
import time
from my_dlib_funcs import *

print(os.getcwd())



# Set working paths
database_path =  os.getcwd() + '/database'
test_path = os.getcwd() + '/testing imgs'
output_path = os.getcwd() +'/Outputs'

# Set models paths
shape_predictor_path = os.getcwd() + '/models/shape_predictor_68_face_landmarks_GTX.dat'
face_recognition_model_path = os.getcwd() + "/models/dlib_face_recognition_resnet_model_v1.dat"
cnn_model_path = os.getcwd() + '/models/mmod_human_face_detector.dat'

# Load the shape predictor and face recognition models

predictor = dlib.shape_predictor(shape_predictor_path)
face_rec = dlib.face_recognition_model_v1(face_recognition_model_path)

# Load the CNN and HOG face detection models

HOG_face_detector = dlib.get_frontal_face_detector()
cnn_face_detector = dlib.cnn_face_detection_model_v1(cnn_model_path)


# get the reference face descriptors info from the people.csv file

filename = 'people.csv'

beg = time.time()
db_face_descriptors = read_db_csv(filename = filename)
print(time.time() - beg)


# display some useful info for debugging

print(len(db_face_descriptors))
print(type(db_face_descriptors))
print(type(db_face_descriptors[0]))

for desc in db_face_descriptors:
    print(desc.keys())
    print(desc["name"])
    print(desc["img path"])

print("********************************************************")


 # define a video capture object
cap = cv2.VideoCapture(0)

width  = int(cap.get(3))  # float `width`
height = int(cap.get(4))  # float `height`

output_file = output_path + '/output__azeaz_.avi'
 # define an output VideoWriter  object
out = cv2.VideoWriter(output_file,
                      cv2.VideoWriter_fourcc(*"MJPG"), 
                      4,(width,height))

# Check if the webcam is opened correctly
if not cap.isOpened():
    print("Error opening video stream or file")

# Read the video frames
while cap.isOpened():
    
    ret, frame = cap.read()

    # If the frame was not read successfully, break the loop
    if not ret:
        print("Error reading frame")
        break
      
    # start recording time (to calculate FPS later)
    print(frame.shape)


    # Capture the video frame
    # by frame
    #frame = cv2.resize(frame,(320,240))
    # Display the resulting frame
    #cv2.imshow('frame', frame)
        
    descriptors = get_face_descriptors(frame  ,                                            
                                       detection_scheme='HOG',                         
                                       shape_predictor = predictor, 
                                       face_recognizer = face_rec , 
                                       upsampling = 1)
    
    recognize(target_descriptors = descriptors,
              database_descriptors = db_face_descriptors, 
              max_dist_thresh = 0.7 )
    
    # get the details for each detected face in the frame i.e bounding boxes and name
    for desc in descriptors:
        print(len(descriptors))
        print(desc["name"])
        
        # get bounding box coordinates
        left = desc["bounding box"][0]
        top = desc["bounding box"][1]
        right = desc["bounding box"][2]
        bottom = desc["bounding box"][3]
        
        # put the face label and bounding box in the final ouput frame
        frame  = cv2.rectangle(frame ,(left,top),(right,bottom),(255,0,0),thickness = 4)
        frame  = cv2.putText(frame , desc["name"], (left - 5 ,top - 5), cv2.FONT_HERSHEY_SIMPLEX, 0.5, (0, 0, 255), 2)

    # display webcam stream with results        
    cv2.imshow("Output", frame )
    
    # calculate FPS
    end = time.time()
    fps = 1/(end - beg)
    out.write(frame)
    print(f'FPS = {fps:.2f}')
        
        # the 'q' button is set as the
        # quitting button you may use any
        # desired button of your choice
    if cv2.waitKey(1) & 0xFF == ord('q'):
        break

# After the loop release the cap object
cap.release()
out.release()
# Destroy all the windows
cv2.destroyAllWindows()