Python人脸识别速度慢

Question

我正在尝试构建一个使用人脸识别库实时检测人脸的软件。我使用网络摄像头进行了尝试，结果令人满意且帧速率相当稳定，但是当我切换到 .mp4 视频时，结果在 fps 方面非常糟糕。我在 OpenCV 中使用 Python 3.6，这是我正在使用的代码：

import face_recognition
import cv2


# Load a sample picture and learn how to recognize it.
totti_image = face_recognition.load_image_file("totti.jpg")
totti_face_encoding = face_recognition.face_encodings(totti_image)[0]

# Create arrays of known face encodings and their names
known_face_encodings = [
    totti_face_encoding
]
known_face_names = [
    "Francesco Totti"
]
def get_faces(frame):
    # Convert the image from BGR color (which OpenCV uses) to RGB color (which face_recognition uses)
    rgb_frame = frame[:, :, ::-1]

    # Find all the faces and face enqcodings in the frame of video
    face_locations = face_recognition.face_locations(rgb_frame)
    face_encodings = face_recognition.face_encodings(rgb_frame, face_locations)

    # Loop through each face in this frame of video
    for (top, right, bottom, left), face_encoding in zip(face_locations, face_encodings):
        # See if the face is a match for the known face(s)
        matches = face_recognition.compare_faces(known_face_encodings, face_encoding, tolerance=0.50)

        name = "Unknown"

        # If a match was found in known_face_encodings, just use the first one.
        if True in matches:
            first_match_index = matches.index(True)
            name = known_face_names[first_match_index]

        # Draw a box around the face
        cv2.rectangle(frame, (left, top), (right, bottom), (0, 0, 255), 2)

        # Draw a label with a name below the face
        cv2.rectangle(frame, (left, bottom - 35), (right, bottom), (0, 0, 255), cv2.FILLED)
        font = cv2.FONT_HERSHEY_DUPLEX
        cv2.putText(frame, name, (left + 6, bottom - 6), font, 1.0, (255, 255, 255), 1)

    return frame

函数“get_faces”在每帧的 while 循环内被调用，我的性能大约为 0.5 fps。如果有人有建议以在输出中获得更好的 fps，请告诉我，谢谢。

编辑： 我使用了以下示例（根据我的需要调整它）并且一切都运行得更好： 链接

最终代码：

import face_recognition
import cv2

# Load a sample picture and learn how to recognize it.
image = face_recognition.load_image_file("totti.jpg")
encoding = face_recognition.face_encodings(image)[0]

# Create arrays of known face encodings and their names
known_face_encodings = [
    encoding
]
known_face_names = [
    "Totti",
]

# Initialize some variables
face_locations = []
face_encodings = []
face_names = []

def get_faces(frame):

    # Resize frame of video to 1/10 size for faster face recognition processing
    small_frame = cv2.resize(frame, (0, 0), fx=0.1, fy=0.1)

    # Convert the image from BGR color (which OpenCV uses) to RGB color (which face_recognition uses)
    rgb_small_frame = small_frame[:, :, ::-1]

    # Find all the faces and face encodings in the current frame of video
    face_locations = face_recognition.face_locations(rgb_small_frame)
    face_encodings = face_recognition.face_encodings(rgb_small_frame, face_locations)

    face_names = []
    for face_encoding in face_encodings:
        # See if the face is a match for the known face(s)
        matches = face_recognition.compare_faces(known_face_encodings, face_encoding)
        name = "Person"

        # If a match was found in known_face_encodings, just use the first one.
        if True in matches:
            first_match_index = matches.index(True)
            name = known_face_names[first_match_index]

        face_names.append(name)


    # Display the results
    for (top, right, bottom, left), name in zip(face_locations, face_names):
        # Scale back up face locations since the frame we detected in was scaled to 1/10 size
        top *= 10
        right *= 10
        bottom *= 10
        left *= 10

        # Draw a box around the face
        cv2.rectangle(frame, (left, top), (right, bottom), (0, 0, 255), 2)

        # Draw a label with a name below the face
        cv2.rectangle(frame, (left, bottom - 35), (right, bottom), (0, 0, 255), cv2.FILLED)
        font = cv2.FONT_HERSHEY_DUPLEX
        cv2.putText(frame, name, (left + 6, bottom - 6), font, 1.0, (255, 255, 255), 1)

    return frame

Answer 1

要确定脚本的哪些部分运行时间最长，请使用分析器。这将输出每个调用的执行时间，因此您可以更好地了解函数的哪些部分是次优的。有关如何分析代码的示例，请参阅Python 分析器。

从文档：

加速人脸识别

如果您的计算机具有多个 CPU 内核，则可以并行执行人脸识别。例如，如果您的系统有 4 个 CPU 内核，则通过并行使用所有 CPU 内核，您可以在相同的时间内处理大约 4 倍的图像。如果您使用 Python 3.4 或更新版本，请传入 --cpus <number_of_cpu_cores_to_use> 参数：

face_recognition --cpus 4 ./pictures_of_people_i_know/ ./unknown_pictures/

您还可以传入 --cpus -1 以使用系统中的所有 CPU 内核。

使用一个内核和最大内核数在您的计算机上测试操作。如果这显着缩短了执行时间，您最好的做法是在您自己的脚本中实现多处理。

更新 2020-08-05

由于它仍然受到一些关注，因此请多研究一下。如果我们看一下存储库，看起来 CLI 只是进行了一些您可以自己编写脚本的调用，以将这些--cpus调用添加到您自己的代码中。具体来说，您可以在此处以编程方式而不是从命令行使用代码。使用多处理以类似方式调用 API，或调用def process_images_in_process_pool(images_to_check, number_of_cpus, model):.