6

I'm trying to create a Python script that detects holes, ends and beginnings of a line. I thought that openCV would be great to achieve this.

So for example everything starts with this image:

enter image description here

finally what I want to achieve is this:

enter image description here

So I began with importing the image into Python and converting it in grayscale. Now I came to the idea to track the holes by using the goodFeaturesToTrack() method. It's normally used to find corners in the image.

However that didn't work so well because after that the script knows the points, but it doesn't know if a point is from a hole or if it's the beginning or end of the line. Another problem is that if I use another image this method detects more points than just the holes, beginnings and ends of the line.

Here is my full code to understand my problem a bit better:

import cv2
import numpy as np
import matplotlib.pyplot as plt

# lodes in img

img = cv2.imread('png1.png', cv2.IMREAD_COLOR)
img_gray = cv2.cvtColor(img, cv2.COLOR_RGB2GRAY)

corners = cv2.goodFeaturesToTrack(img_gray, 200, 0.05, 10)

for corner in corners:
    x, y = corner.ravel()
    cv2.circle(img, (x,y), 7, (255,255,0), -1)

cv2.imshow('img',img)

I have no idea to get around this problem.

jonrsharpe
  • 115,751
  • 26
  • 228
  • 437
RunningRails
  • 63
  • 1
  • 4
  • You could compare the x-y co-ords of each item to the x-y of every other item, and if they fall within a certain threshold of close-ness, consider them a hole, else consider them a start/end line. – Aphire Mar 07 '16 at 15:56
  • 1
    GoodFeaturesToTrack it's not a good option here. I'd do something along this: 1) segment each blob (`findContours`). You'll have each segment. 2) find starting and ending point of each segment (this can be tricky) 3) if two endpoints are closer than a given threshold, then there is a _hole_. 4) The remaining two endpoints are start and end. – Miki Mar 07 '16 at 15:56
  • Ok. Thank You! I did this. Now i need to get all the point in the threshold. Do i need a roi for this or is there a quicker way? – RunningRails Mar 07 '16 at 16:34
  • Looking at your image#2 it's simple: you close the image by radius (~hole tolerance). Then, for any point in the object, the farthest point measured along the object must be one endpoint... Then you find the other endpoint and connected regions not in the original object are your holes. – mainactual Mar 07 '16 at 18:25

1 Answers1

2

I added a func getLandmarks() it returns all the wholes. So here I assume that it will be counted as a hole if there are 2 corners in a radius of 30 pix 

if abs(x1-x2)<=30 and abs(y1-y2)<=30:

This line defines the range.

import cv2
import numpy as np

def getLandmarks(corners):
    holes=[]
    for i in range(0,len(corners)):
        for j in range(i+1,len(corners)):
            x1,y1=corners[i].ravel()
            x2,y2=corners[j].ravel()
            if abs(x1-x2)<=30 and abs(y1-y2)<=30:
                holes.append((int((x1+x2)/2),int((y1+y2)/2)))
    return holes

# lodes in img

img = cv2.imread('img.png', cv2.IMREAD_COLOR)
img_gray = cv2.cvtColor(img, cv2.COLOR_RGB2GRAY)

corners = cv2.goodFeaturesToTrack(img_gray, 200, 0.05, 10)

holes=getLandmarks(corners)
print len(holes)
for corner in holes:
    cv2.circle(img, (corner), 7, (255,255,0), -1)

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

output

enter image description here

Now for the Start and end You can easily sort the corners in either X(if the Path is along left to right) or Y(If the path is along top to down) and the min and max will be your start and end!

Arijit
  • 1,506
  • 14
  • 23