如何用OpenCV Python识别线条与矩形的连接关系？求助解决思路

I’ve tackled similar problems before, so let’s break this down into actionable steps to solve your rectangle/line detection issues and build the connection dictionary you need.

Step 1: Image Preprocessing (Critical for Clean Detection) #

First, we need to reduce noise and highlight edges—this will make both shape and line detection far more reliable.

import cv2
import numpy as np

# Load your image
img = cv2.imread("your_image.png")
gray = cv2.cvtColor(img, cv2.COLOR_BGR2GRAY)

# Reduce noise with Gaussian blur
blurred = cv2.GaussianBlur(gray, (5, 5), 0)

# Detect edges using Canny (tune thresholds based on your image)
edges = cv2.Canny(blurred, threshold1=50, threshold2=150)

Tuning tip: If edges are too sparse, lower threshold1; if there’s too much noise, raise threshold2.

Step 2: Detect Rectangles & Triangles #

Your trouble here likely comes from not filtering contours properly. We’ll use contour approximation to identify polygons, then validate them as rectangles/triangles.

# Find contours (only external ones to ignore nested shapes)
contours, _ = cv2.findContours(edges, cv2.RETR_EXTERNAL, cv2.CHAIN_APPROX_SIMPLE)

shapes = {}
shape_id = 1

for contour in contours:
    # Skip tiny contours (adjust area threshold based on your image)
    area = cv2.contourArea(contour)
    if area < 100:
        continue
    
    # Approximate the contour to a polygon
    epsilon = 0.02 * cv2.arcLength(contour, True)
    approx = cv2.approxPolyDP(contour, epsilon, True)
    
    # Identify shape type
    if len(approx) == 4:
        # Check if it's a rectangle (optional: verify angles are ~90 degrees)
        shape_name = f"rectangle_{shape_id}"
        shapes[shape_name] = {"contour": contour, "points": approx}
        shape_id +=1
    elif len(approx) ==3:
        shape_name = f"triangle_{shape_id}"
        shapes[shape_name] = {"contour": contour, "points": approx}
        shape_id +=1

# Optional: Draw detected shapes to verify
for name, data in shapes.items():
    cv2.drawContours(img, [data["points"]], 0, (0,255,0), 2)
cv2.imshow("Detected Shapes", img)
cv2.waitKey(0)

Key fixes:

• Using cv2.RETR_EXTERNAL ensures we only get outer contours (no inner lines inside shapes).
• The area filter removes tiny noise contours that get misclassified.
• approxPolyDP with a dynamic epsilon (based on contour length) works for shapes of any size.

Step 3: Detect Lines (Including Vertical Ones) #

Hough Line Transform (specifically HoughLinesP for line segments) is the way to go here. We’ll filter lines to focus on vertical ones if needed, but first get all relevant lines.

# Detect line segments
lines = cv2.HoughLinesP(edges, rho=1, theta=np.pi/180, threshold=50, minLineLength=30, maxLineGap=10)

detected_lines = {}
line_id =1

for line in lines:
    x1, y1, x2, y2 = line[0]
    
    # Optional: Filter vertical lines (slope is near infinite)
    # if abs(x2 -x1) < 5:  # Adjust tolerance based on line thickness
    #     continue
    
    # Store line endpoints
    detected_lines[f"line_{line_id}"] = {"start": (x1,y1), "end": (x2,y2)}
    line_id +=1

# Optional: Draw lines to verify
for name, data in detected_lines.items():
    cv2.line(img, data["start"], data["end"], (0,0,255), 2)
cv2.imshow("Detected Lines", img)
cv2.waitKey(0)

Tuning tips:

• minLineLength: Ignore tiny line fragments (adjust based on your image’s line thickness).
• maxLineGap: Allow small gaps in broken lines to be treated as a single line.
• For vertical lines: Check if the x-coordinates of the endpoints are nearly the same (since vertical lines have x1 ≈ x2).

Step 4: Match Lines to Connected Shapes #

Now we’ll use cv2.pointPolygonTest to check if a line’s endpoints lie inside or on the edge of a shape. This tells us which shapes the line connects.

line_shape_map = {}

for line_name, line_data in detected_lines.items():
    connected_shapes = []
    start_point = line_data["start"]
    end_point = line_data["end"]
    
    for shape_name, shape_data in shapes.items():
        # Check if start point is inside/on the shape
        start_test = cv2.pointPolygonTest(shape_data["contour"], start_point, False)
        # Check if end point is inside/on the shape
        end_test = cv2.pointPolygonTest(shape_data["contour"], end_point, False)
        
        if start_test >=0 or end_test >=0:
            connected_shapes.append(shape_name)
    
    line_shape_map[line_name] = connected_shapes

# Print the final dictionary
print("Line to Shape Mapping:")
print(line_shape_map)

How this works:

• pointPolygonTest returns 1 if the point is inside the contour, 0 if on the edge, -1 if outside.
• We only add the shape to the line’s list if either endpoint is inside or touching the shape.

Troubleshooting Common Issues #

• False rectangles/lines: Increase the area threshold for contours, or adjust Canny/Hough parameters to reduce noise.
• Missed shapes: Lower the epsilon value in approxPolyDP (e.g., 0.01 instead of 0.02) to capture more precise polygon approximations.
• Lines not connecting to shapes: Ensure your line detection includes enough of the line segment to reach the shape—increase minLineLength or decrease maxLineGap.

内容的提问来源于stack exchange，提问作者Murli