OpenCV Python: Draw minAreaRect ( RotatedRect not implemented)

Are there any helper methods to draw a rotated rectangle that is returned by cv2.minAreaRect() presumably as ((x1,y1),(x2,y2),angle)? cv2.rectangle() does not support an angle. And since the tuple returned is not of the "RotatedRect" class (because it seems to not be implemented in the Python bindings) there is no points() method, as shown in the C++ tutorial "Creating Bounding rotated boxes and ellipses for contours¶".

How could a rotated rectangle be drawn from lines - rotate about the center point or the first point given?

Solution 1:

rect = cv2.minAreaRect(cnt)
box = cv2.cv.BoxPoints(rect) # cv2.boxPoints(rect) for OpenCV 3.x
box = np.int0(box)
cv2.drawContours(im,[box],0,(0,0,255),2)

should do the trick.

sources:

1) http://opencvpython.blogspot.in/2012/06/contours-2-brotherhood.html

2) Python OpenCV Box2D

Solution 2:

I know this was asked long ago, but I would like to share a different approach as the one proposed by the accepted answer, maybe this could be helpful for someone else (actually this has been done before in C++, but it seems python still lacks of RotatedRect class).

The idea is to define a rotated rectangle from an angle, a size (W and H) and an initial point. This initial point is the relative top-left corner (the top-left corner of the same size rectangle with no rotation angle). From here, the four vertices can be obtained, which allows us to draw the rotated rectangle with four lines.

class RRect:
  def __init__(self, p0, s, ang):
    self.p0 = (int(p0[0]),int(p0[1]))
    (self.W, self.H) = s
    self.ang = ang
    self.p1,self.p2,self.p3 = self.get_verts(p0,s[0],s[1],ang)
    self.verts = [self.p0,self.p1,self.p2,self.p3]

  def get_verts(self, p0, W, H, ang):
    sin = numpy.sin(ang/180*3.14159)
    cos = numpy.cos(ang/180*3.14159)
    P1 = (int(self.H*sin)+p0[0],int(self.H*cos)+p0[1])
    P2 = (int(self.W*cos)+P1[0],int(-self.W*sin)+P1[1])
    P3 = (int(self.W*cos)+p0[0],int(-self.W*sin)+p0[1])
    return [P1,P2,P3]

  def draw(self, image):
    print(self.verts)
    for i in range(len(self.verts)-1):
      cv2.line(image, (self.verts[i][0], self.verts[i][1]), (self.verts[i+1][0],self.verts[i+1][1]), (0,255,0), 2)
    cv2.line(image, (self.verts[3][0], self.verts[3][1]), (self.verts[0][0], self.verts[0][1]), (0,255,0), 2)

(W, H) = (30,60)
ang = 35 #degrees
P0 = (50,50)
rr = RRect(P0,(W,H),ang)
rr.draw(image)
cv2.imshow("Text Detection", image)
cv2.waitKey(200)

I guess, a similar approach can be used to define the rotated rectangle in terms of its center instead of its relative top-left initial point, but I haven't tried it yet.

Solution 3:

Here's a concrete example to draw the rotated rectangle. The idea is to obtain a binary image with Otsu's threshold then find contours using cv2.findContours(). We can obtain the rotated rectangle using cv2.minAreaRect() and the four corner vertices using cv2.boxPoints(). To draw the rectangle we can use cv2.drawContours() or cv2.polylines().

Input -> Output

Code

import cv2
import numpy as np

# Load image, convert to grayscale, Otsu's threshold for binary image
image = cv2.imread('1.jpg')
gray = cv2.cvtColor(image, cv2.COLOR_BGR2GRAY)
thresh = cv2.threshold(gray, 0, 255, cv2.THRESH_BINARY_INV + cv2.THRESH_OTSU)[1]

# Find contours, find rotated rectangle, obtain four verticies, and draw 
cnts = cv2.findContours(thresh, cv2.RETR_EXTERNAL, cv2.CHAIN_APPROX_SIMPLE)
cnts = cnts[0] if len(cnts) == 2 else cnts[1]
rect = cv2.minAreaRect(cnts[0])
box = np.int0(cv2.boxPoints(rect))
cv2.drawContours(image, [box], 0, (36,255,12), 3) # OR
# cv2.polylines(image, [box], True, (36,255,12), 3)

cv2.imshow('image', image)
cv2.waitKey()