我正在开发一个人脸识别项目，但在使用以下代码时遇到问题。它是一个Python程序，使用OpenCV和face_recognition库对图像和实时图像进行人脸识别。我遇到了一些问题，希望能够帮助解决这些问题。当实时时，opencv 工作并打开相机，但是当识别到人脸时，程序直接关闭，这种情况不应该发生，如果有人有任何解决方案，我将不胜感激。
import argparse
import pickle
from collections import Counter
from pathlib import Path
import cv2
import face_recognition
from PIL import Image, ImageDraw
DEFAULT_ENCODINGS_PATH = Path("output/encodings.pkl")
BOUNDING_BOX_COLOR = (255, 0, 0) # Blue color (BGR format)
TEXT_COLOR = (255, 255, 255) # White color (BGR format)
# Create directories if they don't already exist
Path("training").mkdir(exist_ok=True)
Path("output").mkdir(exist_ok=True)
Path("validation").mkdir(exist_ok=True)
parser = argparse.ArgumentParser(description="Recognize faces in an image")
parser.add_argument("--train", action="store_true", help="Train on input data")
parser.add_argument("--validate", action="store_true", help="Validate trained model")
parser.add_argument("--test", action="store_true", help="Test the model with an unknown image")
parser.add_argument("-m", action="store", default="hog", choices=["hog", "cnn"], help="Which model to use for training: hog (CPU), cnn (GPU)")
parser.add_argument("-f", action="store", help="Path to an image with an unknown face")
parser.add_argument("--realtime", action="store_true", help="Enable real-time face detection using the camera")
args = parser.parse_args()
def encode_known_faces(model: str = "hog", encodings_location: Path = DEFAULT_ENCODINGS_PATH) -> None:
"""
Loads images in the training directory and builds a dictionary of their
names and encodings.
"""
names = []
encodings = []
for filepath in Path("training").glob("*/*"):
name = filepath.parent.name
image = face_recognition.load_image_file(filepath)
face_locations = face_recognition.face_locations(image, model=model)
face_encodings = face_recognition.face_encodings(image, face_locations)
for encoding in face_encodings:
names.append(name)
encodings.append(encoding)
name_encodings = {"names": names, "encodings": encodings}
with encodings_location.open(mode="wb") as f:
pickle.dump(name_encodings, f)
def recognize_faces(image_location: str, model: str = "hog", encodings_location: Path = DEFAULT_ENCODINGS_PATH) -> None:
"""
Given an unknown image, get the locations and encodings of any faces and
compares them against the known encodings to find potential matches.
"""
with encodings_location.open(mode="rb") as f:
loaded_encodings = pickle.load(f)
input_image = face_recognition.load_image_file(image_location)
input_face_locations = face_recognition.face_locations(input_image, model=model)
input_face_encodings = face_recognition.face_encodings(input_image, input_face_locations)
pillow_image = Image.fromarray(input_image)
draw = ImageDraw.Draw(pillow_image)
for bounding_box, unknown_encoding in zip(input_face_locations, input_face_encodings):
name = _recognize_face(unknown_encoding, loaded_encodings)
if not name:
name = "Unknown"
_display_face(draw, bounding_box, name)
del draw
pillow_image.show()
def _recognize_face(unknown_encoding, loaded_encodings):
"""
Given an unknown encoding and all known encodings, find the known
encoding with the most matches.
"""
boolean_matches = face_recognition.compare_faces(loaded_encodings["encodings"], unknown_encoding)
votes = Counter(name for match, name in zip(boolean_matches, loaded_encodings["names"]) if match)
if votes:
return votes.most_common(1)[0][0]
def _display_face(draw, bounding_box, name):
"""
Draws bounding boxes around faces, a caption area, and text captions.
"""
top, right, bottom, left = bounding_box
draw.rectangle(((left, top), (right, bottom)), outline=BOUNDING_BOX_COLOR)
text_left, text_top, text_right, text_bottom = draw.textbbox((left, bottom), name)
draw.rectangle(((text_left, text_top), (text_right, text_bottom)), fill=BOUNDING_BOX_COLOR, outline=BOUNDING_BOX_COLOR)
draw.text((text_left, text_top), name, fill=TEXT_COLOR)
def validate(model: str = "hog"):
"""
Runs recognize_faces on a set of images with known faces to validate
known encodings.
"""
for filepath in Path("validation").rglob("*"):
if filepath.is_file():
recognize_faces(image_location=str(filepath.absolute()), model=model)
def detect_faces_realtime(model: str = "hog", encodings_location: Path = DEFAULT_ENCODINGS_PATH) -> None:
"""
Performs real-time face detection using the camera feed.
"""
with encodings_location.open(mode="rb") as f:
loaded_encodings = pickle.load(f)
video_capture = cv2.VideoCapture(0) # Open the camera
while True:
ret, frame = video_capture.read() # Read a frame from the camera feed
# Resize the frame to reduce processing time (optional)
small_frame = cv2.resize(frame, (0, 0), fx=0.25, fy=0.25)
rgb_small_frame = small_frame[:, :, ::-1] # Convert BGR frame to RGB
face_locations = face_recognition.face_locations(rgb_small_frame, model=model)
face_encodings = face_recognition.face_encodings(rgb_small_frame, face_locations)
for (top, right, bottom, left), face_encoding in zip(face_locations, face_encodings):
name = _recognize_face(face_encoding, loaded_encodings)
if not name:
name = "Unknown"
# Scale the face locations back to the original size
top *= 4
right *= 4
bottom *= 4
left *= 4
cv2.rectangle(frame, (left, top), (right, bottom), BOUNDING_BOX_COLOR, 2)
cv2.putText(
frame,
name,
(left + 6, bottom - 6),
cv2.FONT_HERSHEY_SIMPLEX,
1.0,
TEXT_COLOR,
1,
)
cv2.imshow("Video", frame) # Display the resulting frame
if cv2.waitKey(1) & 0xFF == ord("q"):
break # Break the loop if 'q' is pressed
video_capture.release() # Release the camera
cv2.destroyAllWindows() # Close all OpenCV windows
if __name__ == "__main__":
if args.train:
encode_known_faces(model=args.m)
if args.validate:
validate(model=args.m)
if args.test:
recognize_faces(image_location=args.f, model=args.m)
if args.realtime:
detect_faces_realtime(model=args.m)
说明：该代码使用OpenCV和face_recognition库对图像和视频进行人脸识别。它由几个函数组成，包括encode_known_faces、recognize_faces、validate和detect_faces_realtime。
encode_known_faces 函数从训练目录加载图像，检测人脸并提取其编码。然后将这些编码保存在 pickle 文件中。
recognize_faces 函数获取未知图像，检测面部，并将其编码与 pickle 文件中的已知编码进行比较。它显示识别出的面孔以及边界框和标签。
验证函数用于通过在验证目录中具有已知面孔的一组图像上运行 recognize_faces 来验证已知编码。
detector_faces_realtime 函数使用摄像头输入执行实时人脸检测。它连续捕获帧、检测面部并使用已知的编码来识别它们。
该脚本利用 argparse 模块来处理不同功能的命令行参数，例如训练、验证、测试和实时检测。所选模型（hog 或 cnn）也可以指定为命令行参数。
错误代码：
python .\ detector.py --实时 回溯（最近一次调用最后一次）： 文件“C:\Users\Cartu\Downloads\materials-face-recognition\source_code_final\ detector.py”，第 171 行，位于 detector_faces_realtime(模型=args.m) 文件“C:\Users\Cartu\Downloads\materials-face-recognition\source_code_final\ detector.py”，第 130 行，在 detector_faces_realtime 中 face_encodings =face_recognition.face_encodings（rgb_small_frame，face_locations） ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ ^^^^^^^^^^^^^^^ 文件“C:\Python311\Lib\site-packages ace_recognition pi.py”，第 214 行，face_encodings 返回 [np.array(face_encoder.compute_face_descriptor(face_image, raw_landmark_set, num_jitters)) for raw_landmark_set in raw_landmarks] ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ 文件“C:\Python311\Lib\site-packages ace_recognition pi.py”，第 214 行，位于 返回 [np.array(face_encoder.compute_face_descriptor(face_image, raw_landmark_set, num_jitters)) for raw_landmark_set in raw_landmarks] ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ 类型错误：compute_face_descriptor()：函数参数不兼容。支持以下参数类型： 1.（自身：_dlib_pybind11.face_recognition_model_v1，img：numpy.ndarray[（rows，cols，3），numpy.uint8]，脸部：_dlib_pybind11.full_object_detection，num_jitters：int = 0，填充：float = 0.25）-> _dlib_pybind11。向量 2. (self: _dlib_pybind11.face_recognition_model_v1, img: numpy.ndarray[(rows,cols,3),numpy.uint8], num_jitters: int = 0) -> _dlib_pybind11.vector 3.（自身：_dlib_pybind11.face_recognition_model_v1，img：numpy.ndarray[（rows，cols，3），numpy.uint8]，面孔：_dlib_pybind11.full_object_detections，num_jitters：int = 0，填充：float = 0.25）-> _dlib_pybind11。向量 4.（自我：_dlib_pybind11.face_recognition_model_v1，batch_img：列表[numpy.ndarray [（行，列，3），numpy.uint8]]，batch_faces：列表[_dlib_pybind11.full_object_detections]，num_jitters：int = 0，填充：float = 0.25) -> _dlib_pybind11.vectorss 5.（自身：_dlib_pybind11.face_recognition_model_v1，batch_img：列表[numpy.ndarray [（行，列，3），numpy.uint8]]，num_jitters：int = 0）-> _dlib_pybind11.vectors
调用方式：<_dlib_pybind11.face_recognition_model_v1 object at 0x0000019DD1502D70>, array([[[179, 179, 179], [179, 179, 179], [179, 179, 179], ..., [126、129、128]、 [126、130、129]、 [123, 130, 128]],
[[181, 181, 181],
[181, 181, 181],
[180, 180, 180],
...,
[123, 129, 136],
[122, 130, 130],
[122, 131, 131]],
[[180, 180, 180],
[179, 179, 179],
[180, 180, 180],
...,
[122, 128, 130],
[124, 132, 132],
[121, 130, 128]],
...,
[[175, 174, 165],
[172, 173, 168],
[173, 174, 171],
...,
[ 71, 71, 71],
[ 71, 71, 71],
[ 72, 72, 71]],
[[176, 172, 165],
[175, 172, 167],
[172, 173, 169],
...,
[ 65, 67, 67],
[ 68, 69, 69],
[ 60, 64, 63]],
[[176, 171, 169],
[174, 171, 169],
[173, 171, 171],
...,
[ 66, 66, 66],
[ 70, 70, 70],
[ 62, 65, 64]]], dtype=uint8), <_dlib_pybind11.full_object_detection object at 0x0000019DD19F07F0>, 1
我尝试多次更改代码，删除内容以查找错误，我一直在搜索但找不到任何信息。如果有人有其他选择，我看过一些关于 deepface 的内容，但我不知道它是否更容易使用。
简单
from deepface import DeepFace
DeepFace.stream(db_path = r"C:\database", time_threshold=1, frame_threshold=5)
这是我目前得到的最好结果，现在我正在尝试在识别到人脸时发送通知
嘿 cartu28 我还想创建具有人脸识别功能的人体入侵检测，但我想训练模型如何使用 Deepface 来做到这一点。