Made further improvements to the detection algorithm.

camera module/main.py

@@ -3,11 +3,16 @@ import cv2
 # Constant variables definition.
 DESIRED_HEIGHT = 480		# The input image will be resized to this height, preserving its aspect ratio.
 BLUE_THRESHOLD = 150		# If the blue channel is bigger than this, it is considered background and removed.
-BINARY_THRESHOLD = 1		# If the pixel is not brighter than this, it is removed before detection.
+BINARY_THRESHOLD = 30		# If the pixel is not brighter than this, it is removed before detection.
+CANNY_LOW_THRES = 150		# Low threshold for the canny edge detector.
+CANNY_HIGH_THRES = 350		# High threshold for the canny edge detector.
 LINE_THICKNESS = 2			# Thickness of the drawn lines.
+MIN_CONTOUR_AREA = 100		# Min area of a contour to be considered valid.
+MAX_CONTOUR_AREA = 2100		# Max area of a contour to be considered valid.
 BLUR_KERNEL_SIZE = 3		# The size of the Gaussian blur kernel.
 DILATION_KERNEL_SIZE = 5	# The size of the dilation kernel.
 DILATION_ITERATIONS = 5		# The number of dilation iterations.
+MIN_DISTANCE_FOR_MOVE = 10	# Min distance of the drone from the center for the servos to move.
 
 # Colors (assuming the default BGR order).
 RED = (0, 0, 255)
@@ -15,51 +20,85 @@ GREEN = (0, 255, 0)
 BLUE = (255, 0, 0)
 YELLOW = (0, 255, 255)
 
-
 # -------------- Function definitions -----------------------------
 def resizeImage(img):
 	"Resize the input image based on the DESIRED_HEIGHT variable."
-	p = img.shape;
+	p = img.shape
 	aspectRatio = p[0]/p[1]
 	width = DESIRED_HEIGHT*aspectRatio
 	img = cv2.resize(img, ( DESIRED_HEIGHT, int(width) ))
 	return img
 
+def removeColors(img):
+	out = None
+	dim = img.shape
+	blue = img.copy()
+	for i in range(dim[0]):
+		for j in range(dim[1]):
+			pixel = img[i,j]
+			if pixel[0] > 150:
+				blue[i,j,:] = 255
+			else:
+				blue[i,j,:] = 0
+	
+	gray = cv2.cvtColor(blue, cv2.COLOR_BGR2GRAY)
+	_, contours, hierarchy = cv2.findContours(gray, cv2.RETR_EXTERNAL, cv2.CHAIN_APPROX_SIMPLE)
+	if len(contours) > 0:
+		maxContour = max(contours, key = cv2.contourArea)
+		x,y,w,h = cv2.boundingRect(maxContour)
+		out = img[y:y+h,x:x+w,:]
+
+	return out
+
+def findMatchingContour(img, objX, objY):
+
+	dilated = img.copy()
+	#dilated = cv2.dilate(img, (5,5), iterations=1)
+	canny = cv2.Canny(dilated, CANNY_LOW_THRES, CANNY_HIGH_THRES)
+	_, contours, hierarchy = cv2.findContours(canny, cv2.RETR_LIST, cv2.CHAIN_APPROX_NONE)
+	
+	#print('len:' + str(len(contours)))
+	contours.sort(key = cv2.contourArea, reverse = True)
+	
+	cv2.imshow('hey', canny)
+	for i in range(len(contours)):
+		contour = contours[i]
+		x,y,w,h = cv2.boundingRect(contour)
+
+		area = w*h
+		dist = cv2.pointPolygonTest(contour, (objX,objY), False)
+		#print('dist: ' + str(dist))
+		if area >= MIN_CONTOUR_AREA and area <= MAX_CONTOUR_AREA and dist >= 0:
+			return (True, contour)
+
+	return (False, None)
+
 def processImage(img):
 
 	# Resize image to the desired height.
 	resized = resizeImage(img)
-	
+	#return removeColors(resized)
 	dim = resized.shape
 
-	# Remove BLUE
-	noBlue = resized.copy()
-	for i in range(dim[0]):
-		for j in range(dim[1]):
-			if (resized[i,j,0] > BLUE_THRESHOLD):
-				noBlue[i,j,:] = 0
-
 	# Convert to grayscale.
-	gray = cv2.cvtColor(noBlue, cv2.COLOR_BGR2GRAY)
-
+	gray = cv2.cvtColor(resized, cv2.COLOR_BGR2GRAY)
+	
 	# Blur the image.
 	blur = cv2.GaussianBlur(gray, (BLUR_KERNEL_SIZE, BLUR_KERNEL_SIZE), 0)
-	
+
 	# Threshold the image and find its contours.
-	_, imgThres = cv2.threshold(blur, BINARY_THRESHOLD, 255, cv2.THRESH_BINARY)
+	_, imgThres = cv2.threshold(blur, BINARY_THRESHOLD, 255, cv2.THRESH_BINARY_INV)
 	
 	# Dilate the image.
 	dilated = cv2.dilate(imgThres, (DILATION_KERNEL_SIZE,DILATION_KERNEL_SIZE), iterations=DILATION_ITERATIONS)
 	
 	# Find the largest image contour.
-	_, contours, hierarchy = cv2.findContours(dilated, cv2.RETR_TREE, cv2.CHAIN_APPROX_SIMPLE)
-	maxContour = max(contours, key = cv2.contourArea)
-
-	'''
-	hull = cv2.convexHull(maxContour)
-	cv2.drawContours(imgOriginal, maxContour, -1, (0,0,255), 3)
-	cv2.drawContours(imgOriginal, hull, -1, (255,0,0), 3)
-	'''
+	_, contours, hierarchy = cv2.findContours(dilated, cv2.RETR_LIST, cv2.CHAIN_APPROX_SIMPLE)
+	if len(contours) > 0:
+		maxContour = max(contours, key = cv2.contourArea)
+	else:
+		print('No contours found.')
+		return (resized, 0, 0)
 
 	# Get the bounding rectangle of the contour.
 	x,y,w,h = cv2.boundingRect(maxContour)
@@ -69,27 +108,66 @@ def processImage(img):
 	objCenterY = int( (y + y + h) / 2)
 	imgCenterX = int(dim[1]/2)
 	imgCenterY = int(dim[0]/2)
+	#cv2.circle(resized, (objCenterX, objCenterY), 5, BLUE, LINE_THICKNESS)
+	ret, finalContour = findMatchingContour(blur, objCenterX, objCenterY)
+	if (ret == False):
+		return (resized, 0, 0)
 	
+	#cv2.fillPoly(resized, finalContour, BLUE, cv2.LINE_4)
+	x,y,w,h = cv2.boundingRect(finalContour)
+	objCenterX = int( (x + x + w) / 2)
+	objCenterY = int( (y + y + h) / 2)
+
 	# Draw the bounding rectangle  and its centroid to the image.
-	cv2.circle(resized, (objCenterX, objCenterY), 5, YELLOW, LINE_THICKNESS)
+	#cv2.circle(resized, (objCenterX, objCenterY), 5, YELLOW, LINE_THICKNESS)
 	cv2.rectangle(resized, (x,y), (x+w,y+h), RED, LINE_THICKNESS)
-	
-	return resized
+
+	# Determinate the direction of the object relative to the center of the camera.
+	xDir, yDir = determinateDir(imgCenterX, imgCenterY, objCenterX, objCenterY)
+
+	return (resized, xDir, yDir)
+
+def determinateDir(cenX, cenY, objX, objY):
+	xDir = 0
+	yDir = 0
+
+	if abs(cenX - objX) >= MIN_DISTANCE_FOR_MOVE:
+		if objX > cenX:
+			xDir = 1
+		else:
+			xDir = -1
+	if abs(cenY - objY) >= MIN_DISTANCE_FOR_MOVE:
+		if objY > cenY:
+			yDir = -1
+		else:
+			yDir = 1
+	return (xDir, yDir)
 
 #####################################################################################
 
 
-# Read image from source
-img = cv2.imread('/home/stelios/Desktop/IoT/Talos_Drones_Tracking_and_Telemetry/camera module/drone.jpg', cv2.IMREAD_COLOR)
+cap = cv2.VideoCapture('/home/stelios/Desktop/drone_flight_test (cut).mp4')
+if (cap.isOpened() == False):
+	print('Error opening stream.')
+	quit()
+	
+#cap.set(1, 30*6)
+
+while(cap.isOpened()):
+	ret, frame = cap.read()
+	if (ret == True):
+		img, xDir, yDir = processImage(frame)
+		cv2.imshow('Frame', img)
 
-# Process the image and get the output.
-output = processImage(img)
+		print('Got ' + str(xDir) + ' ' + str(yDir))
 
-cv2.imshow('Output', output)
+		k = cv2.waitKey(25) & 0xFF
+		if k == 27:
+			break
+		if k == ord('p') or k == ord('P'):
+			cv2.waitKey(0)
 
-# Terminate if the escape key is pressed or the window is closed.
-while True:
-	k = cv2.waitKey()
-	if k == 27 or k == 255:
+	else:
 		break
+cap.release()
 cv2.destroyAllWindows()