关于OpenCV的坑

  1. 无法修改摄像头参数
  2. 中文无法识别
    2.1 窗口中文标题无法识别
    2.1 中文路径无法识别
  3. 图片转BGR

  4. resize问题

  5. 无法修改摄像头参数
    原因1：摄像头不支持此参数
    比如：摄像头的帧率最高是30，但是你的电脑CPU带不动它，所以它在你电脑上的帧率只能达到22。

原因2：opencv的问题
我查到一个说法是opencv的权限比驱动低，所以无法修改，可以通过卸载摄像头驱动来修复，这个方法没有测试，看着就不靠谱。

因此就是opencv的问题了。
下面这个代码是广泛流传的打开摄像头代码，

  1. import numpy as np
  2. import cv2
  3. cap = cv2.VideoCapture(0) # 打开摄像头
  4. while(True):
  5.     ret, frame = cap.read() # 读取图片
  6.     cv2.imshow('frame',frame)
  7.     # 退出
  8.     if cv2.waitKey(1) & 0xFF == ord('q'):
  9.         break
  10. cap.release()
  11. cv2.destroyAllWindows()

代码是不存在问题，但是打开摄像头之后无法修改修改相机的参数，比如分辨率、帧率、亮度。

  1. cap = cv2.VideoCapture(camera_number + cv2.CAP_DSHOW)

camera_number就是摄像头在你电脑上的设备号，0,1,2,…
CAP_DSHOW是opencv初始化的一个参数，我找到一个参数列表，适用参数应该与相机的驱动有关系，而我使用的笔记本的摄像头（USB相机），因此使用CAP_DSHOW就行了。

  1. CAP_ANY          = 0             # Auto detect
  2. CAP_VFW          = 200           # Video For Windows (obsolete, removed)
  3. CAP_V4L          = 200           # V4L/V4L2 capturing support
  4. CAP_V4L2         = CAP_V4L       # Same as CAP_V4L
  5. CAP_FIREWIRE     = 300           # IEEE 1394 drivers
  6. CAP_FIREWARE     = CAP_FIREWIRE  # Same value as CAP_FIREWIRE
  7. CAP_IEEE1394     = CAP_FIREWIRE  # Same value as CAP_FIREWIRE
  8. CAP_DC1394       = CAP_FIREWIRE  # Same value as CAP_FIREWIRE
  9. CAP_CMU1394      = CAP_FIREWIRE  # Same value as CAP_FIREWIRE
  10. CAP_QT           = 500           # QuickTime (obsolete, removed)
  11. CAP_UNICAP       = 600           # Unicap drivers (obsolete, removed)
  12. CAP_DSHOW        = 700           # DirectShow (via videoInput)
  13. CAP_PVAPI        = 800           # PvAPI, Prosilica GigE SDK
  14. CAP_OPENNI       = 900           # OpenNI (for Kinect)
  15. CAP_OPENNI_ASUS  = 910           # OpenNI (for Asus Xtion)
  16. CAP_ANDROID      = 1000          # Android - not used
  17. CAP_XIAPI        = 1100          # XIMEA Camera API
  18. CAP_AVFOUNDATION = 1200          # AVFoundation framework for iOS (OS X Lion will have the same API)
  19. CAP_GIGANETIX    = 1300          # Smartek Giganetix GigEVisionSDK
  20. CAP_MSMF         = 1400          # Microsoft Media Foundation (via videoInput)
  21. CAP_WINRT        = 1410          # Microsoft Windows Runtime using Media Foundation
  22. CAP_INTELPERC    = 1500          # RealSense (former Intel Perceptual Computing SDK)
  23. CAP_REALSENSE    = 1500          # Synonym for CAP_INTELPERC
  24. CAP_OPENNI2      = 1600          # OpenNI2 (for Kinect)
  25. CAP_OPENNI2_ASUS = 1610          # OpenNI2 (for Asus Xtion and Occipital Structure sensors)
  26. CAP_GPHOTO2      = 1700          # gPhoto2 connection
  27. CAP_GSTREAMER    = 1800          # GStreamer
  28. CAP_FFMPEG       = 1900          # Open and record video file or stream using the FFMPEG library
  29. CAP_IMAGES       = 2000          # OpenCV Image Sequence (e.g. img_%02d.jpg)
  30. CAP_ARAVIS       = 2100          # Aravis SDK
  31. CAP_OPENCV_MJPEG = 2200          # Built-in OpenCV MotionJPEG codec
  32. CAP_INTEL_MFX    = 2300          # Intel MediaSDK
  33. CAP_XINE         = 2400          # XINE engine (Linux)

这是我写的一个完整Demo

  1. import cv2
  2. import os
  4. # Width, Height
  5. Camera_PROP_640x480 = (640, 480)
  6. Camera_PROP_800x600 = (800, 600)
  7. Camera_PROP_960_640 = (960, 640)
  8. Camera_PROP_1024x540 = (1024, 540)
  9. Camera_PROP_1024x600 = (1024, 600)
  10. Camera_PROP_1024x768 = (1024, 768)
  11. Camera_PROP_1080x960 = (1080, 960)
  12. Camera_PROP_1152x864 = (1152, 864)
  13. Camera_PROP_1280x600 = (1280, 600)
  14. Camera_PROP_1280x720 = (1280, 720)
  15. Camera_PROP_1280x768 = (1280, 768)
  16. Camera_PROP_1280x800 = (1280, 800)
  17. Camera_PROP_1280x960 = (1280, 960)
  18. Camera_PROP_1280x1024 = (1280, 1024)
  19. Camera_PROP_1360x768 = (1360, 768)
  20. Camera_PROP_1366x768 = (1366, 768)
  21. Camera_PROP_1400x1050 = (1400, 1050)
  22. Camera_PROP_1440x900 = (1440, 900)
  23. Camera_PROP_1600x900 = (1600, 900)
  24. Camera_PROP_1680x1050 = (1680, 1050)
  25. Camera_PROP_1920x1080 = (1920, 1080)
  26. Camera_PROP_2048x1080 = (2048, 1080)
  27. Camera_PROP_4096x2160 = (4096, 2160)
  28. Camera_PROP_Max = (300000, 300000)
  30. # 解决显示图片过大导致图片跑到显示器外边
  31. def imshow(winname, mat):
  32.     ratio = mat.shape[0] // 600
  33.     if ratio <= 1:
  34.         ratio = 1.5
  35.     cv2.namedWindow(winname, cv2.WINDOW_NORMAL)
  36.     cv2.resizeWindow(winname, int(mat.shape[1] / ratio), int(mat.shape[0] / ratio))
  37.     cv2.imshow(winname, mat)
  39. if __name__ == '__main__':
  40.     print("=============================================")
  41.     print("=  热键(请在摄像头的窗口使用）                   =")
  42.     print("=  w: 拍摄图片                                =")
  43.     print("=  q: 退出                                   =")
  44.     print("=============================================")
  45.     # 新建文件夹 test
  46.     class_name = 'test'
  47.     if not os.path.exists(class_name):
  48.         os.mkdir(class_name)
  49.     index = 1 
  51.     # 打开相机
  52.     cap = cv2.VideoCapture(0 + cv2.CAP_DSHOW)
  53.     # fps = cap.get(cv2.CAP_PROP_FPS)
  55.     # 设置相机亮度
  56.     cap.set(cv2.CAP_PROP_BRIGHTNESS, 150)
  57.     print(cap.get(cv2.CAP_PROP_BRIGHTNESS))
  59.     # 设置相机增益
  60.     cap.set(cv2.CAP_PROP_GAIN, 1)
  61.     print(cap.get(cv2.CAP_PROP_GAIN))
  63.     # 设置相机分辨率
  64.     width, height = Camera_PROP_1280x720
  65.     cap.set(cv2.CAP_PROP_FRAME_WIDTH, width)
  66.     cap.set(cv2.CAP_PROP_FRAME_HEIGHT, height)
  67.     width = int(cap.get(cv2.CAP_PROP_FRAME_WIDTH))
  68.     height = int(cap.get(cv2.CAP_PROP_FRAME_HEIGHT))
  70.     # 初始化opencv显示窗口‘capture’
  71.     cv2.namedWindow('capture', cv2.WINDOW_NORMAL)
  72.     cv2.resizeWindow('capture', width // 2, height // 2)
  74.     # 设置相机帧率
  75.     cap.set(cv2.CAP_PROP_FPS, 30)
  76.     print(cap.get(cv2.CAP_PROP_FPS))
  78.     # 新建一个'test.avi'文件
  79.     fps = 8
  80.     fourcc = cv2.VideoWriter_fourcc(*'XVID')
  81.     out = cv2.VideoWriter('test.avi', fourcc, fps, (width, height))
  83.     # 将相机录制到'test.avi'文件中
  84.     while (int(cv2.getWindowProperty('capture', 0)) != -1):
  85.         ret, frame = cap.read() # 获取当前帧
  86.         if not ret:
  87.             continue
  88.         frame = cv2.flip(frame, 1, dst=None) # 镜像
  89.         out.write(frame) # 写入帧
  90.         imshow("capture", frame) # 显示
  92.         # 判断相机退出
  93.         input = cv2.waitKey(1) & 0xFF
  94.         if input == ord('w'):
  95.             cv2.imwrite("%s/%d.png" % (class_name, index), frame)
  96.             print("%s: %d 张图片" % (class_name, index))
  97.             index += 1
  98.         elif input == ord('q'):
  99.             break
  100.     # 释放内存
  101.     out.release()
  102.     cap.release()
  103.     cv2.destroyAllWindows()
  1. 中文无法识别
    OpenCV的中文乱码问题暂时还无法解决，可以把系统编码修改为UTF8格式来解决这个问题，但是这个方法的代价可能太高了，因此只能使用一些“曲线救国”的办法。

2.1 窗口中文标题无法识别
方法一：改用C++的OpenCV
方法二：改用Python2版本，字符串变为 u’中文’
还是不建议使用中文标题

2.1 中文路径无法识别

  1. import cv2
  2. file_name = '测试/测试.bmp'
  3. img = cv2.imread(file_name)
  4. cv2.imshow('测试',img)
  5. cv2.waitKey()

报错：

  1. cv2.error: OpenCV(4.1.1) C:\projects\opencv-python\opencv\modules\highgui\src\window.cpp:352: error: (-215:Assertion failed) size.width>0 && size.height>0 in function 'cv::imshow'

方法一：
用np.fromfile读取，OpenCV再从内存中读取图片

  1. import cv2
  2. import numpy as np
  3. file_name = '测试/测试.bmp'
  4. img = cv2.imdecode(np.fromfile(file_name, dtype=np.uint8), -1)
  5. cv2.imshow('test',img)
  6. cv2.waitKey()
  7. # 中文路径保存
  8. cv2.imencode('.bmp', img)[1].tofile('测试/测试2.bmp')

方法二：
使用其他库读取图片，然后转成numpy格式再转成OpenCV格式，注意OpenCV是BGR

  1. import cv2
  2. import numpy as np
  3. from PIL import Image
  4. Image.MAX_IMAGE_PIXELS = 1000000000  # 图片最大内存
  5. file_name = '测试/测试.bmp'
  6. img_plt = Image.open(file_name)
  7. # pillow转cv
  8. img_cv = cv2.cvtColor(np.asarray(img_plt), cv2.COLOR_RGB2BGR)
  9. # cv转pillow
  10. img_plt = Image.fromarray(cv2.cvtColor(img_cv, cv2.COLOR_BGR2RGB))
  11. cv2.imshow('test', img_cv)
  12. cv2.waitKey()

方法三：
这个方法跟上一个方法类似，但是我觉得比较有意思，是使用urllib通过url来获取图片，接着转成numpy格式再转为OpenCV格式

  1. import cv2
  2. import numpy as np
  3. from urllib.request import urlopen,pathname2url
  4. file_name = '测试/测试.bmp'
  5. file_url = 'file:' + pathname2url(file_name)
  6. img_url = urlopen(file_url)
  7. img_np = np.asarray(bytearray(img_url.read()), dtype="uint8")
  8. img_cv = cv2.imdecode(img_np, cv2.IMREAD_COLOR)
  9. cv2.imshow('test', img_cv)
  10. cv2.waitKey()

3. 图片转BGR

opencv貌似除了cv2.imread(file_name, 0)外，没有自动匹配位深度转灰度，因此在将图片转灰度需要考虑位深度，否则会报错

  1. def convert_BGR(image):
  2.     """
  3.     转BGR
  4.     :param image:原图像
  5.     """
  6.     if len(image.shape) == 2:
  7.         image = cv2.cvtColor(image, cv2.COLOR_GRAY2BGR)
  8.     elif len(image.shape) == 3:
  9.         if image.shape[2] == 3:
  10.             return image
  11.         elif image.shape[2] == 4:
  12.             image = cv2.cvtColor(image, cv2.COLOR_BGRA2BGR)
  13.         else:
  14.             assert 'image channel too many'
  15.     else:
  16.         assert 'image channel too many'
  17.     return image

4. resize问题

获取图片尺寸用img.shape()，会得到(高，宽，通道数)，单通道图片是(高，宽)
但是得到的resize图片的顺序是(宽，高)

  1. img = cv2.resize(img, (width,height))

如果修改图片尺寸，用OpenCV的imshow显示正常，但是传到其他控件考虑会产生裂图（图片显示不正常）
这是因为OpenCV的显示使用了优化，而其他控件打开图片是遵守width，height满足4的倍数
比如PyQt的Qlabel控件显示图片需要将OpenCV的图片转换成QImage，不仅仅是传图片，还要对图片的尺寸进行调整，代码如下：

  1. showImage = QtGui.QImage(show.data, show.shape[1], show.shape[0], show.shape[1] * show.shape[2], QtGui.QImage.Format_RGB888)

ref
https://blog.csdn.net/sinat_41657218/article/details/119301570