我正在尝试使用相机测量一块陶瓷的尺寸变化

Question

我正在尝试测量一块陶瓷在加热时尺寸的变化，我正在尝试使用相机和边缘检测来做到这一点。为了测试这一点，我首先使用从相机拍摄的图像（下图），并希望稍后过渡到实时视频源。

这是精明的图像，我理想地希望轮廓能够沿着陶瓷件的整个边缘，并且能够测量侧面的长度和面积。canny image

我一直在使用 canny 和其他一些图像处理算法来定义边缘，然后这些将是使用查找并抓取轮廓，然后使用边界框来测量尺寸（以像素为单位）来循环这些轮廓，但这被证明是很困难的。我对此完全陌生，并且主要使用教程，下面的代码显示了我一直在尝试的一些图像预处理。有更好的方法吗？你能在没有边界框的情况下测量轮廓的长度吗？

任何帮助表示赞赏。

图像预处理

img = cv2.imread('buller_ring_test.jpg') 
imgCanny = cv2.Canny(img, 110, 255)
#imggrey = cv2.cvtColor(img, cv2.COLOR_BGR2GRAY) 
#imgThre = cv2.adaptiveThreshold(imggrey, 255, cv2.ADAPTIVE_THRESH_GAUSSIAN_C, cv2.THRESH_BINARY, 11, 2) 
#imgBlur = cv2.GaussianBlur(img, (5, 5), 1) imgCanny = cv2.Canny(img, 110, 255) plt.figure(figsize = (15,15)) plt.imshow(imgCanny)

循环轮廓

#find the total contours
conts = cv2.findContours(imgThre,cv2.RETR_TREE,cv2.CHAIN_APPROX_SIMPLE)
#fine tune contours
conts = imutils.grab_contours(conts)
print (len(conts))

#create empty image with original dimensions
conts_img = np.zeros(img.shape)
plt.imshow(conts_img)

#draw contours onto the rmpty image to isolate contours
conts_img = cv2.drawContours(conts_img, conts, -1, (0,255,0), 2)
plt.figure(figsize = (15,15))
plt.imshow(conts_img)

#draw contours onto the rmpty image to isolate contours
#create the empty image with original dimensions
conts_img = np.zeros(img.shape)

def midPoint(ptA, ptB):
    return ((ptA[0] + ptB[0])/2, (ptA[1] + ptB[1])/2)



#loop over all the contours coordinates
for c in conts:
    #extract box points
    box = cv2.minAreaRect(c)
    box = cv2.boxPoints(box)
    #convert the box points into integer
    box = np.array(box, dtype = int)
    #print(box)
    
    if cv2.contourArea(c) < 10: #anything above this number gets used
        continue

    #draw the contour c (it will loop over all individual c's to make conts_img)
    cv2.drawContours(conts_img, [c], -1, (0,255,0), 2)
    cv2.drawContours(conts_img, [box], -1, (255,255,255), 2)

    print(box)
    for (x,y) in box:
        cv2.circle(conts_img, (x, y), 2, (255, 0, 0), 1)
        print(x,y)
        (tl, tr, br, bl) = box
        
        #calcualte midPoints for top & bottom
        (tlX, trX) = midPoint(tl, tr)
        (brX, blX) = midPoint(bl, br)
        #draw midPoint dots for top & bottom
        cv2.circle(conts_img, (int(tlX), int(trX)), 1, (255, 0, 0), 2)
        cv2.circle(conts_img, (int(brX), int(blX)), 1, (255, 0, 0), 2)
        #calculate the distance based on the midPoints
        dA = dist.euclidean((tlX, trX), (brX, blX)) #width
        #print the size in px on each contour rectangle
        cv2.putText(conts_img, '{:.1f} px'.format(dA), (int(tlX)-40, int(trX)-40),
         cv2.FONT_HERSHEY_SIMPLEX, 0.5, (255, 255, 255), 1)
        
        #calcualte midPoints for left & right
        (tlX, trX) = midPoint(tl, bl)
        (brX, blX) = midPoint(tr, br)
        #draw midPoint dots for left & right
        cv2.circle(conts_img, (int(tlX), int(trX)), 1, (255, 0, 0), 2)
        cv2.circle(conts_img, (int(brX), int(blX)), 1, (255, 0, 0), 2)
        #calculate the distance based on the midPoints
        dB = dist.euclidean((tlX, trX), (brX, blX)) #height
        #print the size in px on each contour rectangle
        cv2.putText(conts_img, '{:.1f} px'.format(dB), (int(brX)-40, int(blX)-40),
         cv2.FONT_HERSHEY_SIMPLEX, 0.5, (255, 255, 255), 1)
        



#show the processed image with the contours printed
plt.figure(figsize = (15,15))
plt.imshow(conts_img)
plt.show()