2021-01-26 08:55:43

어떤 사람의 사진을 단 한 장만 넣어주면, 동영상 내에서 그 사람의 얼굴만 검출하는 것을 만들어봤습니다. 저는 제 사진을 주고, 영상 내에서 제 얼굴만 검출되도록 예제를 만들어봤습니다. face_recognition 라이브러리와 opencv-python 라이브러리를 활용했습니다. 또한 face_recognition 라이브러리를 설치하기 전에, dlib을 GPU 연산이 되도록 설치했습니다. 

 

특정 사람의 얼굴만 검출하는 프로그램은 아래와 같은 알고리즘으로 작동합니다.

1. 입력해 준 사진 속에서 얼굴을 검출합니다.

2. 그 얼굴에서 특성 벡터를 도출합니다.

 

3. 웹캠으로 촬영되는 프레임 내 얼굴을 모두 검출합니다.

4. 검출된 얼굴에서 특성 벡터를 도출합니다. 

 

5. 검출된 얼굴의 특성 벡터와 입력해 준 사진 속 얼굴의 특성 벡터가 서로 비슷한지 거리로 판단합니다.

6. 비슷하다면(도출한 특성 간 거리가 가까우면), 그 사람의 얼굴이라고 판단하고 검출된 얼굴 주변에 빨간색 박스를 그려줄 것입니다. 

 

 

위 알고리즘을 구현한 파이썬 코드는 다음과 같습니다.

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
import cv2
import face_recognition
 
# load your image
image_to_be_matched = face_recognition.load_image_file('my_image.jpg')
name = "Kyohoon Sim"
 
# encoded the loaded image into a feature vector
image_to_be_matched_encoded = face_recognition.face_encodings(image_to_be_matched)[0]
print(image_to_be_matched_encoded)
 
# open webcam
webcam = cv2.VideoCapture(0)
 
if not webcam.isOpened():
    print("Could not open webcam")
    exit()
 
# loop through frames
while webcam.isOpened():
 
    # read frame from webcam
    status, frame = webcam.read()
 
    if not status:
        print("Could not read frame")
        exit()
 
    # face_locations = face_recognition.face_locations(frame) # HoG 기반 얼굴 검출기
    face_locations = face_recognition.face_locations(frame, number_of_times_to_upsample=0, model="cnn"# CNN 기반 얼굴 검출기
 
    for face_location in face_locations:
 
        # Print the location of each face in this image
        top, right, bottom, left = face_location
 
        # You can access the actual face itself like this:
        face_image = frame[top:bottom, left:right]
 
        try:
            face_encoded = face_recognition.face_encodings(face_image)[0]
            result = face_recognition.compare_faces([image_to_be_matched_encoded], face_encoded, 0.5)
 
            if result[0== True:
                cv2.rectangle(frame, (left, top), (right, bottom), (00255), 2)
                Y = top - 10 if top - 10 > 10 else top + 10
                text = name
                cv2.putText(frame, text, (left, Y), cv2.FONT_HERSHEY_SIMPLEX, 0.7, (00255), 2)
        except:
            pass
 
    # display output
    cv2.imshow("detect me", frame)
 
    # press "Q" to stop
    if cv2.waitKey(1& 0xFF == ord('q'):
        break
 
# release resources
webcam.release()
cv2.destroyAllWindows()
cs

 

실행결과 영상입니다. 

 

youtu.be/AkwXXqu670c

 

영상 외에도 가족들을 대상으로 테스트 해봤는데, 간혹 다른 사람을 순간적으로 저로 인식한 경우도 있긴 했지만, 대체적으로는 제 얼굴만 잘 검출해냅니다. 

 

질문 및 지적은 항상 환영입니다. 댓글로 남겨주시면 감사하겠습니다^^

 

관련 글

[ubuntu+python] 얼굴 인식하기 (face_recognition 라이브러리 설치부터 사용까지)  

우분투 18.04에 nvidia 드라이버 설치 3일간 삽질한 후기

dlib GPU 버전 우분투에 설치 중 경험했던 2가지 문제 해결방법