基于OpenCV Python模板匹配提取图纸仪器连线及关联的技术求助

Hey, let's break down how to tackle this problem step by step—since you already have the instrument coordinates, we can build on that to find the connections. Even getting half the links right is totally achievable with these approaches:

一、先预处理图纸，突出线条特征 #

First, we need to isolate the lines from other elements (like text, instrument symbols) in your drawing:

• 灰度化与二值化: Convert the image to grayscale with cv2.cvtColor(img, cv2.COLOR_BGR2GRAY), then use cv2.threshold() or cv2.adaptiveThreshold() to turn lines (usually dark) into white pixels on a black background (invert the binary if needed).
• 去噪: Use cv2.medianBlur() or cv2.GaussianBlur() to filter out small noise dots—this prevents false positives when detecting lines.
• 线条细化（骨架化）: For thicker lines, use morphological operations (e.g., cv2.erode() followed by cv2.dilate()) to reduce lines to single-pixel width. This makes it easier to calculate endpoints and connections later.

二、提取线条（直线与曲线） #

Drawings usually have two types of connections—let's handle both:

1. 直线连线（最常见） #

Use the Probabilistic Hough Line Transform, which gives you line endpoints directly:

lines = cv2.HoughLinesP(
    blurred_binary_img,
    rho=1,                  # 像素分辨率
    theta=np.pi/180,        # 角度分辨率
    threshold=50,           # 检测直线的最小投票数
    minLineLength=100,      # 忽略短噪线
    maxLineGap=10           # 允许断线的最大间隙
)

Each line returns as (x1, y1, x2, y2)—store these endpoints for matching.

2. 曲线连线 #

If your drawing has curved connections, use contour detection instead:

contours, _ = cv2.findContours(binary_img, cv2.RETR_EXTERNAL, cv2.CHAIN_APPROX_SIMPLE)
# 过滤掉属于仪器的轮廓（和已识别的仪器 bounding box 对比）
line_contours = [cnt for cnt in contours if not is_instrument_contour(cnt, instrument_bboxes)]

For each curve contour, extract its start/end points (or the outermost points) to use for matching.

三、匹配线条与仪器关联关系 #

Now we need to link lines to your detected instruments. Assume you have either instrument center coordinates (cx, cy) or bounding boxes (bboxes):

1. 直线匹配 #

• 距离检测: Calculate the distance between each line endpoint and every instrument's center. If both endpoints are within a threshold (e.g., 1/4 of the instrument's bbox size) of two different instruments, mark them as connected.
• Bbox 相交检测: If lines don't end exactly at instruments, check if the line segment intersects the instrument's bbox. Use cv2.pointPolygonTest() to verify if any point on the line lies inside the bbox.

2. 曲线匹配 #

• For curve contours, take the first and last points of the contour (or find the points farthest apart) and apply the same distance/threshold check as straight lines.
• Alternatively, calculate the minimum distance from all contour points to each instrument—if a cluster of points on one end of the curve is close to instrument A, and another cluster on the other end is close to instrument B, they're connected.

四、优化提升准确率（哪怕只识别半数也能做到） #

• 过滤无效线条: Drop lines shorter than a minimum length (e.g., smaller than the instrument's diameter) to avoid noise.
• 去重连接对: Remove duplicate pairs like (A,B) and (B,A) since they represent the same connection.
• 调整阈值: Test with your sample drawings to adjust distance thresholds—start with a value that works for 70% of obvious connections, then tweak for edge cases.

快速代码示例（直线匹配） #

Here's a snippet to tie it all together:

import cv2
import numpy as np

# 你已识别的仪器中心坐标（替换为你的实际数据）
instrument_centers = [(150, 200), (350, 200), (250, 400)]
distance_threshold = 30  # 根据图纸比例调整

# 加载并预处理图像
img = cv2.imread("drawing_sample.jpg")
gray = cv2.cvtColor(img, cv2.COLOR_BGR2GRAY)
_, binary = cv2.threshold(gray, 127, 255, cv2.THRESH_BINARY_INV)
blurred = cv2.medianBlur(binary, 3)

# 检测直线
lines = cv2.HoughLinesP(blurred, 1, np.pi/180, 50, 100, 10)

connections = []
if lines is not None:
    for line in lines:
        x1, y1, x2, y2 = line[0]
        # 找到第一个端点最近的仪器
        closest_a = None
        min_dist_a = float("inf")
        for idx, (cx, cy) in enumerate(instrument_centers):
            dist = np.sqrt((x1 - cx)**2 + (y1 - cy)**2)
            if dist < min_dist_a:
                min_dist_a = dist
                closest_a = idx
        # 找到第二个端点最近的仪器
        closest_b = None
        min_dist_b = float("inf")
        for idx, (cx, cy) in enumerate(instrument_centers):
            dist = np.sqrt((x2 - cx)**2 + (y2 - cy)**2)
            if dist < min_dist_b:
                min_dist_b = dist
                closest_b = idx
        # 验证连接有效性
        if closest_a != closest_b and min_dist_a < distance_threshold and min_dist_b < distance_threshold:
            connections.append(tuple(sorted((closest_a, closest_b))))

# 去重
unique_connections = list(set(connections))
print("识别到的连接关系:", unique_connections)

内容的提问来源于stack exchange，提问作者JimmyFails