countobj.py

import sys
import cv2
import numpy as np

ROOT_NODE = -1

misc = '''
All parameters are chosen to solve the problem in the tests, more general parameters would be useful
'''

authors = '''
Emanuel Di Nardo
Antonio Riviezzo
Liliana Romano
'''

def usage():
	print 'python countobj.py [mode][fidelity][fidelityValue]'
	print '\tmode : video (default) | image | h-help | info'
	print '\tfidelity : activate fidelity range'
	print '\tfidelityValue : [0.0, 1.0] default 0.7'
	print '\tFor help digits h or help'
	print '\tInfo for authors and disclaimer'

if __name__ == '__main__':
	mode = 'video'
	if len(sys.argv) > 1:
		mode = sys.argv[1].lower()
	if mode == 'video':
		camera = cv2.VideoCapture(0)
		
		if not camera.isOpened():
			print 'No video devices'

		mog = cv2.BackgroundSubtractorMOG()

		while camera.grab():
			_, frame = camera.retrieve()
			frame = frame[50:700, 200:1000] # Restrict the a ROI
			#frame = cv2.blur(frame, (13, 13)) # Some defects
			#frame = cv2.bilateralFilter(frame, 15, 100, 10) # Too slow
			frame = cv2.medianBlur(frame, 15) # Preserve edges (if it is applicable)
			framebw = cv2.cvtColor(frame, cv2.COLOR_BGR2GRAY)
			_, framet = cv2.threshold(framebw, 125, 255, cv2.THRESH_BINARY_INV | cv2.THRESH_OTSU) # Thresholding using otsu
			
			framet = cv2.morphologyEx(framet, cv2.MORPH_OPEN, (5, 5)) # Remove survived noise and simplify objects (less inner contours and false contours)
			
			bgs = mog.apply(framet, learningRate=0.05) # This step is useful to stop counting when an object is placed/removed
			
			cv2.imshow('bgs', bgs)
			k = cv2.waitKey(33)

			if (np.count_nonzero(bgs) > 300):
				continue

			frame2 = framet.copy()
			c, h = cv2.findContours(frame2, cv2.RETR_TREE, cv2.CHAIN_APPROX_NONE) # Find contours
			totalContours = 0
			br = []
			for i in xrange(len(c)):
			 	if h[0][i][3] == ROOT_NODE and cv2.contourArea(c[i]) > 50: # Only external contours are useful for counting, small area are removed
			 		totalContours += 1
			 		poly = cv2.approxPolyDP(c[i], 5, True) 
			 		br.append(cv2.boundingRect(poly))
			for b in br:
				cv2.rectangle(frame, (b[0], b[1]), (b[0] + b[2], b[1] + b[3]), (255, 255, 0), 3)
			cv2.imshow('frame', frame)
			k = cv2.waitKey(33)
			if k == ord('q'):
				break
			print 'Total contours: ', totalContours

		camera.release()
	elif mode == 'image':
		fidelity = False
		fidelityValue = .7
		if len(sys.argv) > 2:
			fidelity = bool(sys.argv[2])
		if len(sys.argv) > 3:
			fidelityValue = float(sys.argv[3])

		# For images
		img = cv2.imread('spoons.jpg')
		imgCopy = img.copy()
		img = cv2.medianBlur(img, 15)
		#img = cv2.adaptiveBilateralFilter(img, (5, 5), 150) # Preserve edges
		#img = cv2.blur(img, (3,3))
		img = cv2.cvtColor(img, cv2.COLOR_BGR2GRAY)
		#imgt = cv2.adaptiveThreshold(img, 255, cv2.ADAPTIVE_THRESH_GAUSSIAN_C, cv2.THRESH_BINARY_INV, 3, 10)
		_, imgt = cv2.threshold(img, 125, 255, cv2.THRESH_BINARY_INV | cv2.THRESH_OTSU)
		#_, imgt = cv2.threshold(img, 125, 255, cv2.THRESH_BINARY_INV)
		imgt = cv2.morphologyEx(imgt, cv2.MORPH_OPEN, (5, 5))

		img2 = imgt.copy()
		c, h = cv2.findContours(img2, cv2.RETR_TREE, cv2.CHAIN_APPROX_NONE)
		fidelityRange = 0
		if fidelity:
			maxArea = .0;
			for i in c: # With images it is convenient to know the greater area
				area = cv2.contourArea(i)
				if area > maxArea:
					maxArea = area
			fidelityRange = maxArea - (maxArea * fidelityValue); # If objects have same size it prevents false detection

		totalContours = 0
		
		br = []
		for i in xrange(len(c)):
			if h[0][i][3] == ROOT_NODE and cv2.contourArea(c[i]) >= fidelityRange:
				totalContours += 1
				approx = cv2.approxPolyDP(c[i], 3, True)
				br.append(cv2.boundingRect(approx))
		for b in br:
			cv2.rectangle(imgCopy, (b[0], b[1]), (b[0] + b[2], b[1] + b[3]), (255, 255, 0), 3)
		cv2.imshow('image',imgCopy)
		cv2.waitKey(0)
		print 'Total contours: ', totalContours
	elif mode == 'h' or mode == 'help':
		usage()
	elif mode == 'info':
		print authors
		print misc
	else:
		print 'No algorithm or info chosen'
		usage()