使用OpenCV检测刻度盘上模糊的小三角标识

• 径向灰度扫描法
  借助已知的刻度盘中心，沿全角度径向扫描像素灰度。由于三角是轻微凸起，其位置的灰度会与周围刻度形成明显差异。通过对比相邻径向的灰度变化，定位突变点即为三角位置：

  import cv2
  import numpy as np
  
  center = (cx, cy)  # 已获取的刻度盘中心坐标
  radius = 100  # 刻度盘半径（根据实际调整）
  angle_step = 1  # 扫描步长（单位：度）
  
  max_gray_diff = 0
  target_angle = 0
  prev_gray = 0
  
  for angle in range(0, 360, angle_step):
      rad = np.deg2rad(angle)
      # 计算当前角度下的径向坐标
      x = int(center[0] + radius * np.cos(rad))
      y = int(center[1] + radius * np.sin(rad))
      # 取3x3区域平均灰度，避免单像素噪点干扰
      gray_val = np.mean(cv2.cvtColor(img[y-1:y+2, x-1:x+2], cv2.COLOR_BGR2GRAY))
      
      if angle > 0:
          gray_diff = abs(gray_val - prev_gray)
          if gray_diff > max_gray_diff:
              max_gray_diff = gray_diff
              target_angle = angle
      prev_gray = gray_val
  
  # 转换为三角点坐标
  tri_x = int(center[0] + radius * np.cos(np.deg2rad(target_angle)))
  tri_y = int(center[1] + radius * np.sin(np.deg2rad(target_angle)))
  

• 轮廓极径极值分析
  针对已获取的刻度盘轮廓，将轮廓点转换为以中心为原点的极坐标，计算每个点的极径。三角作为凸起，其极径会显著大于（或小于）周围刻度的轮廓点，找到局部极值点即可定位：

  import numpy as np
  import cv2
  
  contour = ...  # 已获取的刻度盘轮廓
  center = (cx, cy)  # 刻度盘中心
  polar_data = []
  
  # 转换所有轮廓点为极坐标
  for point in contour:
      x, y = point[0]
      dx = x - center[0]
      dy = y - center[1]
      rad = np.sqrt(dx**2 + dy**2)
      angle = np.arctan2(dy, dx)
      polar_data.append((angle, rad))
  
  # 按角度排序，便于连续分析
  polar_data.sort(key=lambda item: item[0])
  
  # 遍历寻找局部极径最大值（凸起对应更大极径）
  tri_x, tri_y = 0, 0
  for i in range(1, len(polar_data)-1):
      prev_rad = polar_data[i-1][1]
      curr_rad = polar_data[i][1]
      next_rad = polar_data[i+1][1]
      
      if curr_rad > prev_rad and curr_rad > next_rad:
          tri_angle = polar_data[i][0]
          tri_x = int(center[0] + curr_rad * np.cos(tri_angle))
          tri_y = int(center[1] + curr_rad * np.sin(tri_angle))
          break
  

• 局部阈值化+连通域筛选
  对刻度盘ROI进行局部阈值处理，放大三角凸起与背景的差异，再通过连通域面积筛选定位三角：

  import cv2
  
  center = (cx, cy)
  radius = 100
  # 提取刻度盘ROI
  roi = img[center[1]-radius:center[1]+radius, center[0]-radius:center[0]+radius]
  gray_roi = cv2.cvtColor(roi, cv2.COLOR_BGR2GRAY)
  # 自适应阈值化，突出微小灰度差异
  thresh = cv2.adaptiveThreshold(gray_roi, 255, cv2.ADAPTIVE_THRESH_GAUSSIAN_C, cv2.THRESH_BINARY_INV, 11, 2)
  # 提取外部连通域
  contours, _ = cv2.findContours(thresh, cv2.RETR_EXTERNAL, cv2.CHAIN_APPROX_SIMPLE)
  
  # 筛选最小有效连通域（三角面积通常远小于刻度）
  min_area = float('inf')
  target_cnt = None
  for cnt in contours:
      area = cv2.contourArea(cnt)
      if 10 < area < min_area:  # 过滤噪点和大轮廓
          min_area = area
          target_cnt = cnt
  
  # 转换为全局坐标
  if target_cnt is not None:
      x, y, w, h = cv2.boundingRect(target_cnt)
      tri_x = center[0] - radius + x + w//2
      tri_y = center[1] - radius + y + h//2
  

内容的提问来源于stack exchange，提问作者Will