arrow_cv

Steps involved:

• Opening the webcam / video file, if provided through sys.argv, using cv2#VideoCapture

• Then, using cv2#inRange, select all the pixels in the custom red color range -> [0, 0, 120] -> [50, 50, 255]
• Then, create a mask of the captured pixels on the image, using cv2#bitwise_and and selection

• To thresholding image, we first convert it to grayscale colorspace, by using cv2#cvtColor

• Next, we'll threshold the image to get a binary image using cv2#threshold

• After that, we could have some tiny islands/holes in our binary image, which we'll remove using cv2#erode and cv2#dilate respectively
• Then, to smooth the edges for better approximation, we will use cv2#medianBlur

• After that, we will find the contours in our image, using cv2#findContours, in which we use SIMPLE approx to save a lot of memory
• And, then we'll iterate through contours to approximate the polygons, using an epsilon value calculated by arcLength of contour
• Now, as we have all those polygons, we've to select only the arrow, so we'll select those having sides = 7 and minArea = 1000.0
• This will possibly give us only the arrow shape, mostly
• Now we'll iterate on each vertex, like

for i in range(7):
	if i == 6: # if it's the last index, take first element
		theta = np.arctan2 (
			pts[i][1] - pts[0][1], # diff of their y-coord 
			pts[i][0] - pts[0][0] # diff of their x-coord
		)

	else: # else take consecutive elements
		theta = np.arctan2 (
			pts[i][1] - pts[i+1][1], 
			pts[i][0] - pts[i+1][0]
		)

	# limit the theta in 0..360 and append it to inclinations
	inc.append((theta * 180 / np.pi + 360) % 360)

• Now, we have inclinations of every lines, let's find the difference

for i in range(7):
	if i == 6: # the same logic as in prev loop
		diff = abs(inc[i] - inc[0]) # diff of first and last
	else:
		diff = abs(inc[i] - inc[i+1]) # diff of consectutive elem

	# append the angles rounded to nearest multiple of 45 deg
	ang.append(round(diff / 45) * 45)

deg = [a % 90 for a in ang] # take the angles in I quad

• Now, we've the side angles too, we've to just find the index of 90 deg in between two 45 deg, as

if deg.count(45) == 2: # if only two 45 deg are there
	f = deg.index(45) # the first 45 deg
	l = 6 - deg[::-1].index(45) # the last 45 deg

	# the following if-else extract the index of 90deg when it's in between and when it's not
	if abs(f-l) == 2: # it's in between
		direc = inc[(f+l)//2]
	elif l == 5: # 90deg is at last
		direc = inc[6]
	elif f == 1: # 90deg is at first
		direc = inc[0]

	direc = ((direc - 45) + 360) % 360 
	# remove the extra 45deg inclination which the arrow head has, and limit it in 0..360

	print(direc) # print IT!!

• That's the whole working I have done, thanks for reading it :)