Python Opencv实现单目标检测的示例代码

def get_roi_contours(image_path, morph_size, num_morph):
  '''
  参数详解：
  image_path：所需处理图片路径
  morph_size：形态学处理核的大小
  num_morph：进行形态学处理的次数
  '''
  image = cv2.imread(image_path, cv2.IMREAD_COLOR)
  #灰度转换
  gray_image = cv2.cvtColor(image, cv2.COLOR_BGR2GRAY)
  #二值化
  threhold, binary_image = cv2.threshold(gray_image, 0, 255, cv2.THRESH_BINARY_INV+cv2.THRESH_OTSU)
  #形态学操作
  kernel = cv2.getStructuringElement(shape=cv2.MORPH_RECT, ksize=morph_size)
  morph_image = cv2.morphologyEx(binary_image, cv2.MORPH_CLOSE, kernel)
  for i in range(num_morph-1):
    morph_image = cv2.morphologyEx(morph_image, cv2.MORPH_CLOSE, kernel)
  #查找轮廓
  contours, hierarchy = cv2.findContours(morph_image, cv2.RETR_EXTERNAL, cv2.CHAIN_APPROX_SIMPLE)
  #选取轮廓面积最大的轮廓
  area = 0
  max_area_index = 0
  for j in range(len(contours)):
    if area < cv2.contourArea(contours[j]):
      max_area_index = j
      area = cv2.contourArea(contours[j])
  rect = cv2.boundingRect(contours[max_area_index])
  return rect

def get_roi_ConCom(image_path, morph_size, num_morph):
  '''
  参数详解：
  image_path：所需处理图片路径
  morph_size：形态学处理核的大小
  num_morph：进行形态学处理的次数
  '''
  image = cv2.imread(image_path, cv2.IMREAD_COLOR)
  #灰度转换
  gray_image = cv2.cvtColor(image, cv2.COLOR_BGR2GRAY)
  #二值化
  threhold, binary_image = cv2.threshold(gray_image, 0, 255, cv2.THRESH_BINARY_INV+cv2.THRESH_OTSU)
  #形态学操作
  kernel = cv2.getStructuringElement(shape=cv2.MORPH_RECT, ksize=morph_size)
  morph_image = cv2.morphologyEx(binary_image, cv2.MORPH_CLOSE, kernel)
  for i in range(num_morph-1):
    morph_image = cv2.morphologyEx(morph_image, cv2.MORPH_CLOSE, kernel)
  #联通组件查询
  numlabels, components_img, stats, centers = cv2.connectedComponentsWithStats(morph_image, 8)
  #获取除背景外的所有联通组件
  stats_without_back = stats[1:]
  #获取除背景外的所有联通组件的面积最大值
  max_area = np.max(stats_without_back, axis=0)[-1]
  #获取面积最大联通组件的index
  max_area_index = stats_without_back[:, -1]==max_area
  rect = stats_without_back[max_area_index]
  return np.squeeze(rect)[0:4]

def get_roi_Kmeans(image_path):
  image = cv2.imread(image_path, cv2.IMREAD_COLOR)
  image_data = image.reshape(-1, 3).astype(np.float32) #必须要转成浮点类型进行计算
  #簇内平方和，标签和每个簇的中心
  criteria = (cv2.TERM_CRITERIA_EPS + cv2.TERM_CRITERIA_COUNT, 10, 1)
  interia, label, centers = cv2.kmeans(image_data, 2, None, criteria, 5, cv2.KMEANS_RANDOM_CENTERS)
  #二值化，将标签为0的转换为255，即是目标
  label[label==0] = 255
  label[label==1] = 0
  #转换数据类型，轮廓查找要是uint8类型数据
  thresh_img = label.reshape(image.shape[0:2]).astype(np.uint8)
  x_grad = cv2.Sobel(thresh_img, cv2.CV_32F, 1, 0)
  y_grad = cv2.Sobel(thresh_img, cv2.CV_32F, 0, 1)

  x_grad = cv2.convertScaleAbs(x_grad) #ax + b 线性变换
  y_grad = cv2.convertScaleAbs(y_grad)
  
  dst = cv2.add(x_grad, y_grad, dtype=cv2.CV_16S) #将两种sobel的加起来就可以得到整个边缘
  dst = cv2.convertScaleAbs(dst)
  plt.imshow(dst, cmap='gray')
  #轮廓查找目标必须为1
  contours, hierarchy = cv2.findContours(dst, cv2.RETR_EXTERNAL, cv2.CHAIN_APPROX_SIMPLE)
  #获取外接矩形
  rect = cv2.boundingRect(contours[0])
  return rect