def test_calibration_transformation_matrix(self):
test_matrix = numpy.zeros((200, 200))
corners = {
'top_left': V(0, 0),
'top_right': V(199, 0),
'bottom_left': V(0, 199),
'bottom_right': V(199, 199)
}
trans = calibration.calibration_transformation_matrix(200, 200, corners)
self.assertEquals(trans[90][90], V(90, 90))
self.assertEquals(trans[0][0], V(0, 0))
self.assertEquals(trans[199][199], V(199, 199))
test_matrix = numpy.zeros((200, 200))
corners = {
'top_left': V(100, 0),
'top_right': V(199, 0),
'bottom_left': V(100, 199),
'bottom_right': V(199, 199)
}
trans = calibration.calibration_transformation_matrix(200, 200, corners)
self.assertEquals(trans[100][100], V(0, 100))
self.assertEquals(trans[1][1], V(-199, 1))
self.assertEquals(trans[199][199], V(199, 199))
def _calibrate(self, white_frame, black_frame):
""" Fill gui with white, capture a frame, fill with black, capture another frame.
Substract the images and calculate a threshold, generate a gradient to get the borders.
Calculate a transformation matrix that converts from the coordinates on the frame to screen coordinates.
"""
height, width = white_frame.shape
# Calculate threshold and gradient to end up with an image with just the border of the screen as white pixels
diff_frame = cv2.subtract(white_frame, black_frame)
threshold_frame = cv2.threshold(diff_frame, settings.CALIBRATION_THRESHOLD, 255, cv2.THRESH_BINARY)[1]
gradient_frame = cv2.Laplacian(threshold_frame, cv2.CV_64F)
gradient_frame = self.threshold(gradient_frame, 255, 256, 255.0)
cv2.imwrite("white.jpg", white_frame)
cv2.imwrite("black.jpg", black_frame)
cv2.imwrite("diff.jpg", diff_frame)
cv2.imwrite("threshold.jpg", threshold_frame)
cv2.imwrite("gradient.jpg", gradient_frame)
self.gui.fill(Color(1, 1, 1))
self.gui.update()
# Get list of all white pixels in the gradient as [(y,x), (y,x), ...]
border_candidate_points = numpy.transpose(gradient_frame.nonzero())
if not border_candidate_points.any():
return
borders = {}
for border in ['left', 'right', 'top', 'bottom']:
borders[border] = {
'count': 0,
'mean': V(0,0),
'vectors': []
}
for x in range(0, 500):
raw_y, raw_x = random.choice(border_candidate_points)
can = V(raw_x, raw_y)
up = down = left = right = None
# Walk along a 13x13 square path around the point and look for other border points
try:
for x in range(-6, 7):
for y in [-6, 7]:
if gradient_frame[can.y+y][can.x+x] == 255:
if y < 0:
down = can + (x,y)
else:
up = can + (x,y)
for y in range(-6, 7):
for x in [-6, 7]:
if gradient_frame[can.y+y][can.x+x] == 255:
if x < 0:
left = can + (x,y)
else:
right = can + (x,y)
except IndexError:
continue
# If two opposing sides of the square have border points, and all three points are roughly on a line
# then assume this is an actual proper point on the screen border
point_found = False
if up and down and not left and not right:
p = up - can
q = V(down.x - can.x, can.y - down.y)
point_found = True
elif left and right and not up and not down:
p = V(can.x-left.x, left.y-can.y)
q = V(right.x-can.x, can.y-right.y)
point_found = True
if point_found:
# Test if the three points are roughly on one line
if p*q/(p.abs()*q.abs()) > 0.95:
# For now, we will simply assume that the image is roughly centered,
# i.e. that the left edge is on the left half of the screen, etc
if up and down:
if can.x < width/2:
border = 'left'
else:
border = 'right'
else:
if can.y < height/2:
border = 'top'
else:
border = 'bottom'
borders[border]['count'] += 1
borders[border]['mean'] += can
borders[border]['vectors'].append(can)
# Paint discovered border point on original black frame for debugging
black_frame[can.y][can.x] = 255
else:
black_frame[can.y][can.x] = 128
cv2.imwrite("debug.jpg", black_frame)
# Go through list of discovered border points and calculate vectors for the borders
self.gui.draw_image_slow(black_frame)
for (border_name, border) in borders.items():
if not border['count']:
self.gui.update()
return
# Divide mean vector by number of vectors to get actual mean for this border
real_mean = border['real_mean'] = border['mean']/border['count']
# Calculate mean direction of border
direction_mean = V(0,0)
direction_count = 0
for vector in border['vectors']:
if not vector:
continue
direction_vector = real_mean - vector
if (direction_vector+direction_mean).abs() < direction_mean.abs():
direction_vector = -direction_vector
direction_mean += direction_vector
direction_count += 1
border['direction_mean'] = direction_mean = direction_mean/direction_count
added = real_mean + direction_mean
colors = {
'left': Color(255, 0, 0),
'right': Color(0, 255, 0),
'top': Color(255, 255, 0),
'bottom': Color(0, 0, 255)
}
self.gui.draw_line(real_mean.x, real_mean.y, added.x, added.y, stroke_color = colors[border_name])
self.gui.draw_circle((real_mean.x, real_mean.y), 5, stroke_color = colors[border_name])
corners = {}
# Calculate corners from border intersections
for border1, border2 in [('top', 'left'), ('top', 'right'), ('bottom', 'left'), ('bottom', 'right')]:
o1 = borders[border1]['real_mean']
d1 = borders[border1]['direction_mean']
o2 = borders[border2]['real_mean']
d2 = borders[border2]['direction_mean']
corner = V.intersection(o1, d1, o2, d2)
if not corner:
self.gui.update()
return
print corner
corners[border1+"_"+border2] = corner
self.gui.draw_circle((corner.x, corner.y), 10, Color(255, 0, 255))
print corners
# Calculate transformation matrix
self.gui.calibration_matrix = calibration.calibration_transformation_matrix(width, height, self.gui.width, self.gui.height, corners)
m = self.gui.calibration_matrix
b = corners['top_left']
p = m[b.y+10][b.x+10]
self.gui.draw_circle((p.x, p.y), 3, stroke_color = Color(0,1,0))
b = corners['top_right']
p = m[b.y+10][b.x-10]
self.gui.draw_circle((p.x, p.y), 3, stroke_color = Color(0,1,0))
b = corners['bottom_left']
p = m[b.y-10][b.x+10]
self.gui.draw_circle((p.x, p.y), 3, stroke_color = Color(0,1,0))
b = corners['bottom_right']
p = m[b.y-10][b.x-10]
self.gui.draw_circle((p.x, p.y), 3, stroke_color = Color(0,1,0))
self.gui.update()
print "Calibration successful."
