如何用OpenCV检测相互交叠的独立矩形并获取其坐标？

需要检测屏幕上随机出现的多个矩形坐标，已知矩形宽度，但用轮廓法检测宽度存在误差。当前代码只能检测到整个矩形块，无法识别出独立的3个矩形，希望获取每个矩形的准确坐标。

现有实时检测代码：

yellow = (5,242,206)
while True:
 isFrameValid, frame = capture.read()

 gray = cv2.cvtColor(frame, cv2.COLOR_BGR2GRAY)
 roi = gray [0:300, 0:1920]

 threshold, thresh_image = cv2.threshold(roi, 30, 255, cv2.THRESH_BINARY)

 # 查找轮廓
 contours, _ =cv2.findContours(thresh_image,cv2.RETR_EXTERNAL,cv2.CHAIN_APPROX_SIMPLE)
 cv2.drawContours(frame, contours, -1,yellow,1) 

问题复现代码：

import cv2

img= cv2.imread('./IwOXW.png')

yellow = (5,242,206)

gray = cv2.cvtColor(img, cv2.COLOR_BGR2GRAY)
roi = gray [0:300, 0:1920]

threshold, thresh_image = cv2.threshold(roi, 30, 255, cv2.THRESH_BINARY)

# 查找轮廓
contours, _ = cv2.findContours(thresh_image,cv2.RETR_EXTERNAL,cv2.CHAIN_APPROX_SIMPLE)
cv2.drawContours(img, contours, -1,yellow,1) 

cv2.imshow('Frame',img)
cv2.waitKey(0)  

示例图片说明：画面中有3个垂直排列的深色矩形，位于浅色背景上，矩形之间存在细小的间隙。

1. 替换固定阈值为自适应阈值 #

固定阈值容易受全局明暗影响，导致矩形间的背景被误判为前景，使轮廓粘连。改用自适应阈值能根据局部区域调整阈值，更精准分离前景和背景：

# 替换原固定阈值代码
thresh_image = cv2.adaptiveThreshold(
    roi, 255, cv2.ADAPTIVE_THRESH_GAUSSIAN_C, cv2.THRESH_BINARY_INV, 11, 2
)

使用THRESH_BINARY_INV是因为目标矩形为深色，反转后矩形变为白色前景，更便于轮廓检测。

2. 形态学开运算分离粘连轮廓 #

若阈值调整后仍有轮廓粘连，用开运算（先腐蚀再膨胀）断开连通区域，参数可根据矩形间隙大小调整：

import numpy as np

# 定义结构元素，尺寸根据实际间隙调整
kernel = np.ones((3, 3), np.uint8)
# 开运算分离粘连
thresh_image = cv2.morphologyEx(thresh_image, cv2.MORPH_OPEN, kernel)

3. 基于已知尺寸筛选目标轮廓 #

利用已知的矩形宽度（及高度）范围，过滤掉不符合的干扰轮廓，确保只保留目标矩形：

# 遍历所有轮廓，筛选符合尺寸的矩形
for cnt in contours:
    x, y, w, h = cv2.boundingRect(cnt)
    # 替换为你实际的宽度、高度范围
    w_min, w_max = 45, 55
    h_min, h_max = 190, 210
    if w_min < w < w_max and h_min < h < h_max:
        # 绘制矩形并打印坐标
        cv2.rectangle(img, (x, y), (x + w, y + h), yellow, 2)
        print(f"矩形坐标：左上角({x}, {y})，右下角({x+w}, {y+h})")

完整优化后的复现代码 #

import cv2
import numpy as np

img = cv2.imread('./IwOXW.png')
yellow = (5,242,206)

gray = cv2.cvtColor(img, cv2.COLOR_BGR2GRAY)
roi = gray[0:300, 0:1920]

# 自适应阈值处理
thresh_image = cv2.adaptiveThreshold(
    roi, 255, cv2.ADAPTIVE_THRESH_GAUSSIAN_C, cv2.THRESH_BINARY_INV, 11, 2
)

# 形态学开运算分离粘连轮廓
kernel = np.ones((3, 3), np.uint8)
thresh_image = cv2.morphologyEx(thresh_image, cv2.MORPH_OPEN, kernel)

# 查找轮廓
contours, _ = cv2.findContours(thresh_image, cv2.RETR_EXTERNAL, cv2.CHAIN_APPROX_SIMPLE)

# 筛选并标记目标矩形
for cnt in contours:
    x, y, w, h = cv2.boundingRect(cnt)
    # 根据实际尺寸调整范围
    if 45 < w < 55 and 190 < h < 210:
        cv2.rectangle(img, (x, y), (x + w, y + h), yellow, 2)
        print(f"矩形坐标：左上角({x}, {y})，右下角({x+w}, {y+h})")

cv2.imshow('Frame', img)
cv2.waitKey(0)
cv2.destroyAllWindows()

内容的提问来源于stack exchange，提问作者Seriousl Lancerl