How to find the orientation of an object (shape)? - Python Opencv

Question

My images are always like this:

But I need to rotate them to be like this:

But to do that, I need to find the orientation of the object, knowing that the thinner part of the object has to be on the left side. In summary, the images are wings and the start of the wing has to be on the left side and the end of the wing has to be on the right side.

I hope someone can give me a suggestion, I've tried a bunch of different strategies but with no good result so far.

Answer 1

I used this python library for oriented object detection many times for similar tasks. This is a trainable neural network detecting the positions and orientations of objects. When you know the orientation angle, you can rotate the object to the desired angle using opencv. You will need to label some images to train the network.

Answer 2

Here is one way in Python/OpenCV.

• Read the image

• Convert to grayscale

• Threshold

• Get outer contour

• Get minAreaRect points and angle from outer contour

• Get vertices of rotated rectangle

• Draw the rotated rectangle

• Correct the angle as needed

• Print the angle

• Save the image with the rotated rectangle drawn on it

Input:

import cv2
import numpy as np

# load image as HSV and select saturation
img = cv2.imread("wing2.png")
hh, ww, cc = img.shape

# convert to gray
gray = cv2.cvtColor(img, cv2.COLOR_BGR2GRAY)

# threshold the grayscale image
ret, thresh = cv2.threshold(gray,0,255,0)

# find outer contour
cntrs = cv2.findContours(thresh, cv2.RETR_EXTERNAL, cv2.CHAIN_APPROX_SIMPLE)
cntrs = cntrs[0] if len(cntrs) == 2 else cntrs[1]

# get rotated rectangle from outer contour
rotrect = cv2.minAreaRect(cntrs[0])
box = cv2.boxPoints(rotrect)
box = np.int0(box)

# draw rotated rectangle on copy of img as result
result = img.copy()
cv2.drawContours(result,[box],0,(0,0,255),2)

# get angle from rotated rectangle
angle = rotrect[-1]

# from https://www.pyimagesearch.com/2017/02/20/text-skew-correction-opencv-python/
# the `cv2.minAreaRect` function returns values in the
# range [-90, 0); as the rectangle rotates clockwise the
# returned angle trends to 0 -- in this special case we
# need to add 90 degrees to the angle
if angle < -45:
    angle = -(90 + angle)
 
# otherwise, just take the inverse of the angle to make
# it positive
else:
    angle = -angle

print(angle,"deg")

# write result to disk
cv2.imwrite("wing2_rotrect.png", result)

cv2.imshow("THRESH", thresh)
cv2.imshow("RESULT", result)
cv2.waitKey(0)
cv2.destroyAllWindows()

Angle Returned: 0.8814040422439575 deg

Image with Rotated Rectangle:

Answer 3

You have to compute the main axis (it is based on PCA). It will give you a good idea of the main orientation, then you can rotate your image accordingly.

As it was pointed as a comment, you now have to test that the thin part is on the right side of the image, and for that you use the centroid/barycenter: if the centroid is on the left of the bounding box, then the wing is well oriented.

Here is the complete algorithm:

• Compute the main orientation (main axis, computed with PCA)
• Rotate your image according to the main axis orientation.
• Compute the bounding box and centroid/barycenter
• If the centroid is on the left, then your image is well oriented, else rotate it about 180°.

Here are the results...