import cv2

import sys

import os

import matplotlib

import numpy as np

from collections import Counter

import face_recognition

import dlib

from math import hypot

from object_detection import yoloV3Detect

from landmark_models import *

from face_spoofing import *

from headpose_estimation import *

from face_detection import get_face_detector, find_faces

l = os.listdir('student_db')

known_face_encodings = []

known_face_names = []

face_locations = []

face_encodings = []

face_names = []

for image in l:

obama_image = face_recognition.load_image_file('student_db/'+image)

obama_face_encoding = face_recognition.face_encodings(obama_image)[0]

known_face_encodings.append(obama_face_encoding)

known_face_names.append(image.split('.')[0])

h_model = load_hp_model('models/Headpose_customARC_ZoomShiftNoise.hdf5')

face_model = get_face_detector()

predictor = dlib.shape_predictor("models/shape_predictor_68_face_landmarks.dat")

video_capture = cv2.VideoCapture(0)

process_this_frame = False

no_of_frames_0 = 0

no_of_frames_1 = 0

no_of_frames_2 = 0

no_of_frames_3 = 0

no_of_frames_4 = 0

no_of_frames_5 = 0

no_of_frames_6 = 0

no_of_frames_7 = 0

font = cv2.FONT_HERSHEY_PLAIN

flag = True

def alert(condition,no_of_frames):

return no_of_frames

while True:

# frame skipping to save time

process_this_frame = not process_this_frame

# Grab a single frame of video

ret, frame = video_capture.read()

frame2 = frame.copy()

frame3 = frame.copy()

report = np.zeros((frame3.shape[0],frame3.shape[1], 3), np.uint8)

# Resize frame of video to 1/4 size for faster face recognition processing

small_frame = cv2.resize(frame, (0, 0), fx=0.25, fy=0.25)

# Functionalities

if process_this_frame:

try:

##### Object Detection #####

try:

fboxes,fclasses=yoloV3Detect(small_frame)

to_detect=['person','laptop','cell phone','book','tv']

temp1,temp2=[],[]

for i in range(len(fclasses)):

if(fclasses[i] in to_detect):

temp1.append(fboxes[i])

temp2.append(fclasses[i])

# Conter

count_items = Counter(temp2)

except Exception as e:

count_items = {}

count_items['person'] = 0

count_items['laptop'] = 0

count_items['cell phone'] = 0

count_items['book'] = 0

count_items['tv'] = 0

print(e)

# Multiple Person Buffer

condition = (count_items['person']!=1)

no_of_frames_0 = alert(condition,no_of_frames_0)

y_pos = 20

alert_pos = (120,190)

# Display

cv2.putText(report, "Number of people detected: "+str(count_items['person']), (1, y_pos), font, 1.1, (0, 255, 0), 2)

# Alert

if(no_of_frames_0>10):

cv2.putText(report, "Number of people detected: "+str(count_items['person']), (1, y_pos), font, 1.1, (0, 0, 255), 2)

cv2.putText(report, "ALERT", alert_pos, font, 4, (0, 0, 255), 2)

# Object Detection Buffer

condition = (count_items['laptop']>=1 or

count_items['cell phone']>=1 or

count_items['book']>=1 or

count_items['tv']>=1)

no_of_frames_1 = alert(condition,no_of_frames_1)

# Display

cv2.putText(report, "Banned objects detected: "+str(condition), (1, y_pos+20), font, 1.1, (0, 255, 0), 2)

# Alert

if(no_of_frames_1>10):

cv2.putText(report, "Banned objects detected: "+str(condition), (1, y_pos+20), font, 1.1, (0, 0, 255), 2)

cv2.putText(report, "ALERT", alert_pos, font, 4, (0, 0, 255), 2)

if(count_items['person']==1):

#### face detection using caffe model of OpenCV's DNN module ####

# detect face

faces = find_faces(small_frame, face_model)

if len(faces) >0:

face = faces[0]

else:

condition = (len(faces) < 1)

no_of_frames_7 = alert(condition,no_of_frames_7)

y_pos = 60

alert_pos = (120,190)

# Display

cv2.putText(report, "Number of face detected: "+str(len(faces)), (1, y_pos), font, 1.1, (0, 255, 0), 2)

# Alert

if(no_of_frames_7>10):

cv2.putText(report, "Number of face detected: "+str(len(faces)), (1, y_pos), font, 1.1, (0, 0, 255), 2)

cv2.putText(report, "ALERT", alert_pos, font, 4, (0, 0, 255), 2)

horizontalAppendedImg = np.hstack((frame3,report))

cv2.imshow("Proctoring_Window", horizontalAppendedImg)

continue

# Display Face Detection

(left, top,right,bottom) = face

cv2.rectangle(frame3, (left*4, top*4), (right*4, bottom*4), (0, 0, 255), 2)

if(flag==True):

#### face verification using face_recognition library ####

# modifying order

face_locations = [[top, right, bottom, left]]

# Convert BGR image to RGB image (which uses)

rgb_small_frame = small_frame[:, :, ::-1]

# get CNN feature vector

face_encodings = face_recognition.face_encodings(rgb_small_frame, face_locations)

# get similarity

face_encoding = face_encodings[0]

matches = face_recognition.compare_faces(known_face_encodings, face_encoding)

face_distances = face_recognition.face_distance(known_face_encodings, face_encoding)

best_match_index = np.argmin(face_distances)

if matches[best_match_index]:

name = known_face_names[best_match_index]

else:

name = "Unknown"

flag = False

# Buffer

condition = (name=='Unknown')

no_of_frames_2 = alert(condition,no_of_frames_2)

# Display

cv2.putText(report, "Face Recognized: "+str(name), (1, y_pos+40), font, 1.1, (0, 255, 0), 2)

# Alert

if(no_of_frames_2>10):

cv2.putText(report, "Face Recognized: "+str(name), (1, y_pos+40), font, 1.1, (0, 0, 255), 2)

cv2.putText(report, "ALERT", alert_pos, font, 4, (0, 0, 255), 2)

#### mouth movement ####

# get landmarks

left = face[0]*4

top = face[1]*4

right = face[2]*4

bottom = face[3]*4

face_dlib = dlib.rectangle(left, top, right, bottom)

gray = cv2.cvtColor(frame, cv2.COLOR_BGR2GRAY)

facial_landmarks = predictor(gray, face_dlib)

mouth_ratio = get_mouth_ratio([60,62,64,66], frame2,facial_landmarks)

# Buffer

condition = (mouth_ratio>0.1)

no_of_frames_4 = alert(condition,no_of_frames_4)

# Display

cv2.putText(report, "Mouth Open: "+str(condition), (1, y_pos+80), font, 1.1, (0, 255, 0), 2)

# Alert

if(no_of_frames_4>10):

cv2.putText(report, "Mouth Open: "+str(condition), (1, y_pos+80), font, 1.1, (0, 0, 255), 2)

cv2.putText(report, "ALERT", alert_pos, font, 4, (0, 0, 255), 2)

#### head pose ####

oAnglesNp,oBboxExpanded = headpose_inference(h_model, frame2, face)

# Display (head angle)

frame3 = displayHeadpose(frame3, oAnglesNp,oOffset = 0)

# Buffer

condition1=(round(oAnglesNp[0],1) not in [0.0,-1.0,-1.1,-1.2,-1.3,-1.4,-1.5,-1.6,-1.7] and

round(oAnglesNp[1],0) not in [0.0,1.0,2.0,3.0,4.0,5.0])

no_of_frames_5 = alert(condition1,no_of_frames_5)

# Display

if(condition1):

cv2.putText(report, "Head Pose: Looking away from screen", (1, y_pos+100), font, 1.1, (0, 255, 0), 2)

else:

cv2.putText(report, "Head Pose: Looking at screen", (1, y_pos+100), font, 1.1, (0, 255, 0), 2)

# Alert

if(no_of_frames_5>10):

cv2.putText(report, "Head Pose: Looking away from screen", (1, y_pos+100), font, 1.1, (0, 0, 255), 2)

cv2.putText(report, "ALERT", alert_pos, font, 4, (0, 0, 255), 2)

##### Blinking (to support down eye tracking) ######

left_eye_ratio = get_blinking_ratio([36, 37, 38, 39, 40, 41], frame2,facial_landmarks)

right_eye_ratio = get_blinking_ratio([42, 43, 44, 45, 46, 47], frame2,facial_landmarks)

blinking_ratio = (left_eye_ratio + right_eye_ratio) / 2

##### eye tracker #####

gaze_ratio1_left_eye, gaze_ratio2_left_eye = get_gaze_ratio([36, 37, 38, 39, 40, 41], frame2,facial_landmarks)

gaze_ratio1_right_eye, gaze_ratio2_right_eye = get_gaze_ratio([42, 43, 44, 45, 46, 47], frame2,facial_landmarks)

# Left/Right

new_frame1 = np.zeros((500, 500, 3), np.uint8)

gaze_ratio1 = (gaze_ratio1_right_eye + gaze_ratio1_left_eye) / 2

if gaze_ratio1 <= 0.35:

new_frame1[:] = (0, 0, 255) # red, right

elif 0.35 < gaze_ratio1 < 4:

new_frame1[:] = (0, 0, 0) #black, center

else:

new_frame1[:] = (255, 0, 0) # blue, left

# Up/Down

new_frame2 = np.zeros((500, 500, 3), np.uint8)

# Buffer

condition = (gaze_ratio1 <= 0.35 or gaze_ratio1>=4 or condition1==True)

no_of_frames_3 = alert(condition,no_of_frames_3)

# Display

if(condition):

cv2.putText(report, "Eye Tracking: Looking away from screen", (1, y_pos+60), font, 1.1, (0, 255, 0), 2)

else:

cv2.putText(report, "Eye Tracking: Looking at screen", (1, y_pos+60), font, 1.1, (0, 255, 0), 2)

# Alert

if(no_of_frames_3>10):

cv2.putText(report, "Eye Tracking: Looking away from screen", (1, y_pos+60), font, 1.1, (0, 0, 255), 2)

cv2.putText(report, "ALERT", alert_pos, font, 4, (0, 0, 255), 2)

#### face spoofing ####

measures = face_spoof(frame2,face)

# Buffer

condition = (np.mean(measures) < 0.7)

no_of_frames_6 = alert(condition,no_of_frames_6)

# Display

cv2.putText(report, "Spoof Face detected: "+str(condition), (1, y_pos+120), font, 1.1, (0, 255, 0), 2)

# Alert

if(no_of_frames_6>10):

cv2.putText(report, "Spoof Face detected: "+str(condition), (1, y_pos+120), font, 1.1, (0, 0, 255), 2)

cv2.putText(report, "ALERT", alert_pos, font, 4, (0, 0, 255), 2)

else:

flag = True

horizontalAppendedImg = np.hstack((frame3,report))

cv2.imshow("Proctoring_Window", horizontalAppendedImg)

except Exception as e:

print(e)

flag = True

report = np.zeros((frame3.shape[0],frame3.shape[1], 3), np.uint8)

#final display frame

horizontalAppendedImg = np.hstack((frame3,report))

cv2.imshow("Proctoring_Window", horizontalAppendedImg)

# Display the resulting image

# Hit 'q' on the keyboard to quit!

if cv2.waitKey(1) & 0xFF == ord('q'):

print("closing window...")

break

video_capture.release()

cv2.destroyAllWindows()