python opencv形状检测

Question

For a school project I'm trying to recognize a hexagon shape in a video capture using python and opencv. The problem is that when I use this code, it never finds a hexagon. I haven't found a reason why it isn't working so I hope someone can help me.

The code:

import numpy as np
import cv2

stop_cascade = cv2.CascadeClassifier('cascade.xml')

cap = cv2.VideoCapture(0)

while(True):
    ret, img = cap.read()
    lower = np.array([0,0,0])
    upper = np.array([20,20,20])
    mask = cv2.inRange(img, lower, upper)
    contours, h = cv2.findContours(mask, 1, 2)
    contours.sort(key = len)

    for contour in contours[-3:]:
        approx = cv2.approxPolyDP(contour,0.1*cv2.arcLength(contour,True),True)
        if len(approx) == 6:
            print "hexagon"

    cv2.imshow('image', mask)

    if cv2.waitKey(1) & 0xFF == ord('q'):
        break

cap.release()
cv2.destroyAllWindows()

Thanks in advance

Edit: To be a little more specific. I've tried mulitple different hexagons but when I print the len(approx) it is never higher then 4. Is this a fault in my object or is it that I have an error in finding the contours i don't know about?

Answer 1

Looks like your program is simplifying the contour beyond what you would like. So approxPolyDP is further simplifying your hexagon to a square or triangle. This would also explain why len(approx) never returns a value greater than 4. Try changing the value of epsilon that you pass to approxPolyDP . These are the changes I suggest:

1. Try a range of values for epsilon in approxPolyDP . Change the multiplication factor in 0.1*cv2.arcLength(contour,True) from 0.01 to 0.1
2. Use drawContours to actually display the approx curve. This will help you choose the best multiplication factor

Another approach is using the convex hull of the hexagon rather than the contour and then apply the approxPolyDP function. Ideally, if your image is sharp enough just checking the number of points in the convex hull should be enough to tell you whether the shape is a hexagon or not. Simplifying the hull further should help you be more robust to noisy images. But again choosing the correct multiplication factor will be important.

Answer 2

Use this:

approx = cv2.approxPolyDP(contour,0.03*cv2.arcLength(contour,True),True) 

you will get the desired result.