def __init__(self, frame):
self.blob_image = opencv.cvCloneImage(frame.iplimage)
self.blob_gray = opencv.cvCreateImage(opencv.cvGetSize(self.blob_image), 8, 1)
self.blob_mask = opencv.cvCreateImage(opencv.cvGetSize(self.blob_image), 8, 1)
opencv.cvSet(self.blob_mask, 1)
opencv.cvCvtColor(self.blob_image, self.blob_gray, opencv.CV_BGR2GRAY)
# opencv.cvEqualizeHist(self.blob_gray, self.blob_gray)
# opencv.cvThreshold(self.blob_gray, self.blob_gray, frame.thresh, 255, opencv.CV_THRESH_BINARY)
# opencv.cvThreshold(self.blob_gray, self.blob_gray, frame.thresh, 255, opencv.CV_THRESH_TOZERO)
opencv.cvThreshold(self.blob_gray, self.blob_gray, frame.bthresh, 255, frame.bthreshmode)
# opencv.cvAdaptiveThreshold(self.blob_gray, self.blob_gray, 255, opencv.CV_ADAPTIVE_THRESH_MEAN_C, opencv.CV_THRESH_BINARY_INV)
self._frame_blobs = CBlobResult(self.blob_gray, self.blob_mask, 100, True)
self._frame_blobs.filter_blobs(10, 10000)
self.count = self._frame_blobs.GetNumBlobs()
self.items = []
for i in range(self.count):
self.items.append(self._frame_blobs.GetBlob(i))
class Blobs:
def __init__(self, frame):
self.blob_image = opencv.cvCloneImage(frame.iplimage)
self.blob_gray = opencv.cvCreateImage(opencv.cvGetSize(self.blob_image), 8, 1)
self.blob_mask = opencv.cvCreateImage(opencv.cvGetSize(self.blob_image), 8, 1)
opencv.cvSet(self.blob_mask, 1)
opencv.cvCvtColor(self.blob_image, self.blob_gray, opencv.CV_BGR2GRAY)
#opencv.cvEqualizeHist(self.blob_gray, self.blob_gray)
#opencv.cvThreshold(self.blob_gray, self.blob_gray, frame.thresh, 255, opencv.CV_THRESH_BINARY)
#opencv.cvThreshold(self.blob_gray, self.blob_gray, frame.thresh, 255, opencv.CV_THRESH_TOZERO)
opencv.cvThreshold(self.blob_gray, self.blob_gray, frame.bthresh, 255, frame.bthreshmode)
#opencv.cvAdaptiveThreshold(self.blob_gray, self.blob_gray, 255, opencv.CV_ADAPTIVE_THRESH_MEAN_C, opencv.CV_THRESH_BINARY_INV)
self._frame_blobs = CBlobResult(self.blob_gray, self.blob_mask, 100, True)
self._frame_blobs.filter_blobs(10, 10000)
self.count = self._frame_blobs.GetNumBlobs()
self.items = []
for i in range(self.count):
self.items.append(self._frame_blobs.GetBlob(i))
frame = highgui.cvQueryFrame(capture)
if frame is None:
# no image captured... end the processing
break
# mirror the captured image
cv.cvFlip(frame, None, 1)
cv.cvCvtColor(frame, my_grayscale, cv.CV_RGB2GRAY)
cv.cvThreshold(my_grayscale, my_grayscale, 128, 255,
cv.CV_THRESH_BINARY)
if not blob_overlay:
# Convert black-and-white version back into three-color representation
cv.cvCvtColor(my_grayscale, frame, cv.CV_GRAY2RGB)
myblobs = CBlobResult(my_grayscale, mask, 100, True)
myblobs.filter_blobs(10, 10000)
blob_count = myblobs.GetNumBlobs()
for i in range(blob_count):
my_enumerated_blob = myblobs.GetBlob(i)
# print "%d: Area = %d" % (i, my_enumerated_blob.Area())
my_enumerated_blob.FillBlob(frame, hsv2rgb(i * 180.0 / blob_count),
0, 0)
# we can now display the images
highgui.cvShowImage('Blob View', frame)
# handle events
def main():
if len(sys.argv) == 1:
print 'Usage: %s [inputfile]' % sys.argv[0]
sys.exit(1)
# initialize window
cvNamedWindow('threshold', CV_WINDOW_AUTOSIZE)
cvMoveWindow('threshold', 10, 10)
cvNamedWindow('combined', CV_WINDOW_AUTOSIZE)
cvMoveWindow('combined', 10, 10)
cvNamedWindow('video', CV_WINDOW_AUTOSIZE)
cvMoveWindow('video', 10, 10)
cvNamedWindow('flow', CV_WINDOW_AUTOSIZE)
cvMoveWindow('flow', 500, 10)
cvNamedWindow('edges', CV_WINDOW_AUTOSIZE)
cvMoveWindow('edges', 500, 10)
capture = cvCreateFileCapture(sys.argv[1])
if not capture: print 'Error opening capture'; sys.exit(1)
# Load bg image
bg = cvLoadImage('bg.png')
# Discard some frames
cvSetCaptureProperty(capture, CV_CAP_PROP_POS_FRAMES,
#8276399)
7919999)
#for i in xrange(2300):
# cvGrabFrame(capture)
#print cvGetCaptureProperty(capture,
# CV_CAP_PROP_POS_FRAMES)
frame = cvQueryFrame(capture)
frame_size = cvGetSize(frame)
# vars for playback
fps = 25
play = True
velx = cvCreateImage(frame_size, IPL_DEPTH_32F, 1)
vely = cvCreateImage(frame_size, IPL_DEPTH_32F, 1)
combined = cvCreateImage(frame_size, IPL_DEPTH_8U, 1)
prev = cvCreateImage(frame_size, IPL_DEPTH_8U, 1)
curr = cvCreateImage(frame_size, IPL_DEPTH_8U, 1)
frame_sub = cvCreateImage(frame_size, IPL_DEPTH_8U, 3)
flow = cvCreateImage(frame_size, IPL_DEPTH_8U, 3)
edges = cvCreateImage(frame_size, IPL_DEPTH_8U, 1)
prev_edges = None
storage = cvCreateMemStorage(0)
blob_mask = cvCreateImage (frame_size, IPL_DEPTH_8U, 1)
cvSet(blob_mask, 1)
hough_in = cvCreateImage(frame_size, IPL_DEPTH_8U, 1)
hough_storage = cvCreateMat( 100, 1, CV_32FC3 )
seg_mask = cvCreateImage(frame_size, IPL_DEPTH_32F, 1)
seg_storage = cvCreateMemStorage(0)
kf_near = cvCreateKalman(4, 2, 0)
kf_far = cvCreateKalman(4, 2, 0)
kf_ball = cvCreateKalman(4, 2, 0)
for kalman in [kf_near, kf_far, kf_ball]:
# transition matrix F is initialized to identity
# want F = [1 0 1 0
# 0 1 0 1
# 0 0 1 0
# 0 0 0 1]
# because x1 = x coordinate and x3 = d/dt x1
# x2 = y coordinate and x4 = d/dt x3
kalman.transition_matrix[0,2] = 1
kalman.transition_matrix[1,3] = 1
cvSetIdentity(kalman.measurement_matrix, 1)
cvSetIdentity(kalman.process_noise_cov, 1e-5)
cvSetIdentity(kalman.measurement_noise_cov, 1e-1)
cvSetIdentity(kalman.error_cov_post, 1)
cvSetIdentity(kf_ball.process_noise_cov, 1e-3)
# initialize random number generator
rng = cvRNG()
# x_k = state variables
x_ball = cvCreateMat(4, 1, CV_32FC1)
cvZero(x_ball)
cvZero(kf_ball.state_post)
x_near = cvCreateMat(4, 1, CV_32FC1)
cvZero(x_near)
cvZero(kf_near.state_post)
# w_k = noise
w_k = cvCreateMat(4, 1, CV_32FC1)
# z_k = measurements
z_k = cvCreateMat(2, 1, CV_32FC1)
frame_count = 0
ball_count = 0
present_count = 0
present = []
while frame_count<585:
if play:
frame_count += 1
frame = cvQueryFrame(capture)
start_time=time.clock()
cvSub(frame, bg, frame_sub)
'''#detect people
found = list(cvHOGDetectMultiScale(frame, storage, win_stride=(8,8),
padding=(32,32), scale=1.05, group_threshold=2))
for r in found:
(rx, ry), (rw, rh) = r
tl = (rx + int(rw*0.1), ry + int(rh*0.07))
br = (rx + int(rw*0.9), ry + int(rh*0.87))
cvRectangle(frame, tl, br, (0, 255, 0), 3)
'''
'''
#color thresholding
hsv = cvCreateImage(frame_size, IPL_DEPTH_8U, 3);
cvCvtColor(frame, hsv, CV_BGR2HSV)
mask = cvCreateMat(frame_size.height, frame_size.width,
CV_8UC1)
cvInRangeS(hsv, (0.03*256, 0.2*256, 0.6*256, 0),
(0.16*256, 1.0*256, 1.0*256, 0), mask)
cvShowImage('threshold', mask)
'''
#optical flow method
# store previous frame
prev, curr = curr, prev
# convert next frame to single channel grayscale
cvCvtColor(frame_sub, curr, CV_BGR2GRAY)
#cvCalcOpticalFlowLK(prev, curr, (3,3), velx, vely)
#cvThreshold(velx, velx, 8.0, 0, CV_THRESH_TOZERO)
cvCalcOpticalFlowHS(prev, curr, 1, velx, vely, 0.5,
(CV_TERMCRIT_ITER, 10, 0))
cvThreshold(velx, velx, 0.5, 0, CV_THRESH_TOZERO)
cvThreshold(vely, vely, 0.5, 0, CV_THRESH_TOZERO)
cvErode(vely, vely,
cvCreateStructuringElementEx(2, 2, 0, 0,
CV_SHAPE_ELLIPSE))
cvAdd(vely, velx, vely)
cvShowImage('flow', vely)
#edge detection/dilation
cvCanny(curr, edges, 50, 100)
cvDilate(edges, edges,
cvCreateStructuringElementEx(7, 7, 0, 0,
CV_SHAPE_ELLIPSE))
#cvErode(edges, edges)
cvShowImage('edges', edges)
# do optical flow using edge detection/dilation
if prev_edges:
cvCalcOpticalFlowHS(prev_edges, edges, 1, velx, vely, 0.5,
(CV_TERMCRIT_ITER, 10, 0))
#cvThreshold(vely, vely, 0.5, 0, CV_THRESH_TOZERO)
#cvShowImage('flow', vely)
prev_edges = edges
cvThreshold(vely, combined, 0.5, 255, CV_THRESH_BINARY)
cvMin(combined, edges, combined)
cvShowImage('combined', combined)
# blobs
myblobs = CBlobResult(edges, blob_mask, 100, True)
myblobs.filter_blobs(10, 10000)
blob_count = myblobs.GetNumBlobs()
near_player = []
far_player = []
possible_ball = []
for i in range(blob_count):
blob = myblobs.GetBlob(i)
# Group blobs by their center point
x, y = blob_center(blob)
if y > 142:
near_player.append(blob)
elif y < 40:
far_player.append(blob)
else:
possible_ball.append(blob)
'''
# Remove outliers - this was commented out because
# it's not as effective as hoped
outliers = [remove_outlier(near_player),
remove_outlier(far_player)]
outliers = [x for x in outliers if x]
print outliers
# If no ball found in the range y=(40, 145) then
# find the closest blob to our state estimate
# that is small enough to be a ball.
if len(possible_ball) == 0:
outliers = [blob for blob in outliers if blob.Area() < 100]
if len(outliers) > 0:
center = cvKalmanPredict(kf_ball)
outliers.sort(key=lambda x: \
distance_sq(blob_center(x), center))
possible_ball.append(outliers[0])
'''
ball = None
maxv = -1
# Find the ball that's moving the fastest
for blob in possible_ball:
# get optical flow velocity at blob center
x, y = blob_center(blob)
x, y = int(x), int(y)
# v = magnitude of velocity vector
v = pow(cvGetReal2D(velx, y, x), 2) + \
pow(cvGetReal2D(vely, y, x), 2)
if v > maxv:
ball = blob
maxv = v
if ball and maxv > 0.01:
x, y = blob_center(ball)
print x, y
#cvRectangle(frame, (ball.MinX(), ball.MinY()),
# (ball.MaxX(), ball.MaxY()), CV_RGB(0, 255, 0))
cvRectangle(frame, (x-5,y-5), (x+5,y+5), CV_RGB(0, 255, 0))
ball_count += 1
#ball.FillBlob(frame, cvScalar(0, 255, 0), 0, 0)
# Ask user if moving ball is present
has_ball = raw_input("Moving ball present [y/N]?")
has_ball = (has_ball.lower() == 'y')
if has_ball: present_count += 1
present.append(has_ball)
print 'ball found in %d of %d frames' % \
(ball_count, present_count)
cvShowImage('video', frame)
#print time.clock()-start_time
''' crashes
#hough transform on dilated image
cvCopy(edges, hough_in)
cvSmooth(hough_in, hough_in, CV_GAUSSIAN, 15, 15, 0, 0)
cvHoughCircles(hough_in, hough_storage, CV_HOUGH_GRADIENT,
4, frame_size[1]/10, 100, 40, 0, 0)
print hough_storage
'''
k = cvWaitKey(1000/fps)
if k == 27: # ESC key
break
elif k == 'p': # play/pause
play = not play
cv2.cv.Not(thresh_image_ipl, inverted_license_plate_grayscale_ipl)
# Min blob size and Max blob size
min_blob_size = 100 # Blob must be 30 px by 30 px
max_blob_size = 10000
threshold = 100
# Plate area as % of image area:
max_plate_to_image_ratio = 0.3
min_plate_to_image_ratio = 0.01
image_area = license_plate_size[0] * license_plate_size[1]
# Mask - Blob extracted where mask is set to 1
# Third parameter is threshold value to apply prior to blob detection
# Boolean indicating whether we find moments
myblobs = CBlobResult(thresh_image_ipl, mask, threshold, True)
myblobs.filter_blobs(min_blob_size, max_blob_size)
blob_count = myblobs.GetNumBlobs()
print "Found %d blob[s] betweeen size %d and %d using threshold %d" % (
blob_count, min_blob_size, max_blob_size, threshold)
for i in range(blob_count):
my_enumerated_blob = myblobs.GetBlob(i)
# print "%d: Area = %d" % (i, my_enumerated_blob.Area())
my_enumerated_blob.FillBlob(
license_plate_grayscale_ipl,
#license_plate_ipl,
#cv2.cv.Scalar(255, 0, 0),
cv2.cv.CV_RGB(255, 0, 0),
0, 0)
cv2.cv.Not(thresh_image_ipl, inverted_license_plate_grayscale_ipl)
# Min blob size and Max blob size
min_blob_size = 100 # Blob must be 30 px by 30 px
max_blob_size = 10000
threshold = 100
# Plate area as % of image area:
max_plate_to_image_ratio = 0.3
min_plate_to_image_ratio = 0.01
image_area = license_plate_size[0] * license_plate_size[1]
# Mask - Blob extracted where mask is set to 1
# Third parameter is threshold value to apply prior to blob detection
# Boolean indicating whether we find moments
myblobs = CBlobResult(thresh_image_ipl, mask, threshold, True)
myblobs.filter_blobs(min_blob_size, max_blob_size)
blob_count = myblobs.GetNumBlobs()
print "Found %d blob[s] betweeen size %d and %d using threshold %d" % (
blob_count, min_blob_size, max_blob_size, threshold)
for i in range(blob_count):
my_enumerated_blob = myblobs.GetBlob(i)
# print "%d: Area = %d" % (i, my_enumerated_blob.Area())
my_enumerated_blob.FillBlob(
license_plate_grayscale_ipl,
#license_plate_ipl,
#cv2.cv.Scalar(255, 0, 0),
cv2.cv.CV_RGB(255, 0, 0),
0,
def main():
if len(sys.argv) == 1:
print 'Usage: %s [inputfile]' % sys.argv[0]
sys.exit(1)
# initialize window
cv.NamedWindow('video', cv.CV_WINDOW_AUTOSIZE)
cv.MoveWindow('video', 10, 10)
cv.NamedWindow('threshold', cv.CV_WINDOW_AUTOSIZE)
cv.MoveWindow('threshold', 10, 500)
cv.NamedWindow('flow', cv.CV_WINDOW_AUTOSIZE)
cv.MoveWindow('flow', 500, 10)
cv.NamedWindow('edges', cv.CV_WINDOW_AUTOSIZE)
cv.MoveWindow('edges', 500, 500)
cv.NamedWindow('combined', cv.CV_WINDOW_AUTOSIZE)
cv.MoveWindow('combined', 1000, 10)
capture = cv.CreateFileCapture(sys.argv[1])
if not capture: print 'Error opening capture'; sys.exit(1)
# Load bg image
bg = cv.LoadImage('bg.png')
# Discard some frames
for i in xrange(2300):
cv.GrabFrame(capture)
frame = cv.QueryFrame(capture)
frame_size = cv.GetSize(frame)
# vars for playback
fps = 25
play = True
velx = cv.CreateImage(frame_size, cv.IPL_DEPTH_32F, 1)
vely = cv.CreateImage(frame_size, cv.IPL_DEPTH_32F, 1)
combined = cv.CreateImage(frame_size, cv.IPL_DEPTH_8U, 1)
prev = cv.CreateImage(frame_size, cv.IPL_DEPTH_8U, 1)
curr = cv.CreateImage(frame_size, cv.IPL_DEPTH_8U, 1)
frame_sub = cv.CreateImage(frame_size, cv.IPL_DEPTH_8U, 3)
edges = cv.CreateImage(frame_size, cv.IPL_DEPTH_8U, 1)
prev_edges = None
storage = cv.CreateMemStorage(0)
blob_mask = cv0.cvCreateImage (frame_size, cv.IPL_DEPTH_8U, 1)
cv0.cvSet(blob_mask, 1)
hough_in = cv.CreateImage(frame_size, cv.IPL_DEPTH_8U, 1)
hough_storage = cv.CreateMat( 100, 1, cv.CV_32FC3 )
'''
cv.CvtColor(frame, prev, cv.CV_BGR2GRAY)
frame = cv.QueryFrame(capture)
cv.CvtColor(frame, curr, cv.CV_BGR2GRAY)
# winSize can't have even numbers
cv.CalcOpticalFlowLK(prev, curr, (3,3), velx, vely)
cv.ShowImage('video', frame)
cv.ShowImage('flow', velx)
cv.WaitKey(0)
'''
while True:
if play:
frame = cv.QueryFrame(capture)
cv.Sub(frame, bg, frame_sub)
'''#detect people
found = list(cv.HOGDetectMultiScale(frame, storage, win_stride=(8,8),
padding=(32,32), scale=1.05, group_threshold=2))
for r in found:
(rx, ry), (rw, rh) = r
tl = (rx + int(rw*0.1), ry + int(rh*0.07))
br = (rx + int(rw*0.9), ry + int(rh*0.87))
cv.Rectangle(frame, tl, br, (0, 255, 0), 3)
'''
#color thresholding
hsv = cv.CreateImage(frame_size, cv.IPL_DEPTH_8U, 3);
cv.CvtColor(frame, hsv, cv.CV_BGR2HSV)
mask = cv.CreateMat(frame_size[1], frame_size[0],
cv.CV_8UC1)
cv.InRangeS(hsv, (0.06*256, 0.2*256, 0.6*256, 0),
(0.16*256, 1.0*256, 1.0*256, 0), mask)
cv.ShowImage('threshold', mask)
#optical flow method
# store previous frame
prev, curr = curr, prev
# convert next frame to single channel grayscale
cv.CvtColor(frame_sub, curr, cv.CV_BGR2GRAY)
#cv.CalcOpticalFlowLK(prev, curr, (3,3), velx, vely)
#cv.Threshold(velx, velx, 8.0, 0, cv.CV_THRESH_TOZERO)
cv.CalcOpticalFlowHS(prev, curr, 1, velx, vely, 0.5,
(cv.CV_TERMCRIT_ITER, 10, 0))
cv.Threshold(velx, velx, 0.5, 0, cv.CV_THRESH_TOZERO)
cv.Threshold(vely, vely, 0.5, 0, cv.CV_THRESH_TOZERO)
cv.Erode(vely, vely,
cv.CreateStructuringElementEx(2, 2, 0, 0,
cv.CV_SHAPE_ELLIPSE))
cv.Add(vely, velx, vely)
cv.ShowImage('flow', vely)
#edge detection
cv.Canny(curr, edges, 50, 100)
cv.Dilate(edges, edges,
cv.CreateStructuringElementEx(7, 7, 0, 0,
cv.CV_SHAPE_ELLIPSE))
cv.ShowImage('edges', edges)
if prev_edges:
cv.CalcOpticalFlowHS(prev_edges, edges, 1, velx, vely, 0.5,
(cv.CV_TERMCRIT_ITER, 10, 0))
cv.Threshold(velx, velx, 0.5, 0, cv.CV_THRESH_TOZERO)
cv.Threshold(vely, vely, 0.5, 0, cv.CV_THRESH_TOZERO)
cv.ShowImage('flow', vely)
prev_edges = edges
cv.Threshold(vely, combined, 0.5, 255, cv.CV_THRESH_BINARY)
cv.Min(combined, edges, combined)
cv.ShowImage('combined', combined)
# blobs
myblobs = CBlobResult(edges, blob_mask, 100, False)
myblobs.filter_blobs(10, 10000)
blob_count = myblobs.GetNumBlobs()
for i in range(blob_count):
my_enumerated_blob = myblobs.GetBlob(i)
# print "%d: Area = %d" % (i, my_enumerated_blob.Area())
my_enumerated_blob.FillBlob(frame,
hsv2rgb( i*180.0/blob_count), 0, 0)
cv.ShowImage('video', frame)
''' crashes
#hough transform on dilated image
#http://wiki.elphel.com/index.php?
# title=OpenCV_Tennis_balls_recognizing_tutorial&redirect=no
cv.Copy(edges, hough_in)
cv.Smooth(hough_in, hough_in, cv.CV_GAUSSIAN, 15, 15, 0, 0)
cv.HoughCircles(hough_in, hough_storage, cv.CV_HOUGH_GRADIENT,
4, frame_size[1]/10, 100, 40, 0, 0)
print hough_storage
'''
k = cv.WaitKey(1000/fps)
if k == 27: # ESC key
break
elif k == 'p': # play/pause
play = not play
def main():
if len(sys.argv) == 1:
print 'Usage: %s [inputfile]' % sys.argv[0]
sys.exit(1)
# initialize window
cvNamedWindow('video', CV_WINDOW_AUTOSIZE)
cvMoveWindow('video', 10, 10)
cvNamedWindow('segment', CV_WINDOW_AUTOSIZE)
cvMoveWindow('segment', 10, 500)
cvNamedWindow('flow', CV_WINDOW_AUTOSIZE)
cvMoveWindow('flow', 500, 10)
cvNamedWindow('edges', CV_WINDOW_AUTOSIZE)
cvMoveWindow('edges', 500, 500)
capture = cvCreateFileCapture(sys.argv[1])
if not capture:
print 'Error opening capture'
sys.exit(1)
# Load bg image
bg = cvLoadImage('bg.png')
# Discard some frames
for i in xrange(2300):
cvGrabFrame(capture)
frame = cvQueryFrame(capture)
frame_size = cvGetSize(frame)
# vars for playback
fps = 25
play = True
velx = cvCreateImage(frame_size, IPL_DEPTH_32F, 1)
vely = cvCreateImage(frame_size, IPL_DEPTH_32F, 1)
combined = cvCreateImage(frame_size, IPL_DEPTH_8U, 1)
prev = cvCreateImage(frame_size, IPL_DEPTH_8U, 1)
frame_gray = cvCreateImage(frame_size, IPL_DEPTH_8U, 1)
frame_sub = cvCreateImage(frame_size, IPL_DEPTH_8U, 3)
flow = cvCreateImage(frame_size, IPL_DEPTH_8U, 3)
edges = cvCreateImage(frame_size, IPL_DEPTH_8U, 1)
prev_edges = cvCreateImage(frame_size, IPL_DEPTH_8U, 1)
storage = cvCreateMemStorage(0)
blob_mask = cvCreateImage(frame_size, IPL_DEPTH_8U, 1)
cvSet(blob_mask, 1)
hough_in = cvCreateImage(frame_size, IPL_DEPTH_8U, 1)
hough_storage = cvCreateMat(100, 1, CV_32FC3)
silhouette = cvCreateImage(frame_size, IPL_DEPTH_8U, 1)
mhi = cvCreateImage(frame_size, IPL_DEPTH_32F, 1)
seg_mask = cvCreateImage(frame_size, IPL_DEPTH_32F, 1)
seg_storage = cvCreateMemStorage(0)
kf_near = cvCreateKalman(4, 2, 0)
kf_far = cvCreateKalman(4, 2, 0)
kf_ball = cvCreateKalman(4, 2, 0)
for kalman in [kf_near, kf_far, kf_ball]:
# transition matrix F is initialized to identity
# want F = [1 0 1 0
# 0 1 0 1
# 0 0 1 0
# 0 0 0 1]
# because x1 = x coordinate and x3 = d/dt x1
# x2 = y coordinate and x4 = d/dt x3
kalman.transition_matrix[0, 2] = 1
kalman.transition_matrix[1, 3] = 1
cvSetIdentity(kalman.measurement_matrix, 1)
cvSetIdentity(kalman.process_noise_cov, 1e-5)
cvSetIdentity(kalman.measurement_noise_cov, 1e-1)
cvSetIdentity(kalman.error_cov_post, 1)
cvSetIdentity(kf_ball.process_noise_cov, 1e-3)
# initialize random number generator
rng = cvRNG()
# x_k = state variables
x_ball = cvCreateMat(4, 1, CV_32FC1)
cvZero(x_ball)
cvZero(kf_ball.state_post)
x_near = cvCreateMat(4, 1, CV_32FC1)
cvZero(x_near)
cvZero(kf_near.state_post)
# w_k = noise
w_k = cvCreateMat(4, 1, CV_32FC1)
# z_k = measurements
z_k = cvCreateMat(2, 1, CV_32FC1)
ball_last = None
while True:
if play:
frame = cvQueryFrame(capture)
# do background subtraction using manually created frame
# TODO do averaging to generate bg
cvSub(frame, bg, frame_sub)
# convert to grayscale
cvCvtColor(frame_sub, frame_gray, CV_BGR2GRAY)
# edge detection/dilation
cvCanny(frame_gray, edges, 50, 100)
cvDilate(
edges, edges,
cvCreateStructuringElementEx(7, 7, 0, 0, CV_SHAPE_ELLIPSE))
#cvErode(edges, edges)
cvShowImage('edges', edges)
# do optical flow using edge detection/dilation image
#cvCalcOpticalFlowLK(prev_edges, edges, (7,7), velx, vely)
cvCalcOpticalFlowHS(prev_edges, edges, 1, velx, vely, 0.5,
(CV_TERMCRIT_ITER, 10, 0))
#cvThreshold(vely, vely, 0.5, 0, CV_THRESH_TOZERO)
#cvAdd(velx, vely, vely)
cvShowImage('flow', vely)
cvCopy(edges, prev_edges)
'''
# segment motion
cvThreshold(vely, silhouette, 0.5, 255, CV_THRESH_BINARY)
cvUpdateMotionHistory(silhouette, mhi, time.clock()/1000, 1)
components = cvSegmentMotion(mhi, seg_mask, seg_storage,
time.clock()/1000,
1/fps)
#http://opencv.willowgarage.com/documentation/python/imgproc_miscellaneous_image_transformations.html?highlight=connectedcomp#CvConnectedComp
for c in components:
area, value, rect = c.area, c.value, c.rect
x, y, w, h = rect.x, rect.y, rect.width, rect.height
cvRectangle(frame, (x,y), (x+w,y+h), CV_RGB(255, 0, 0))
cvShowImage('segment', seg_mask)
'''
blobs = CBlobResult(edges, blob_mask, 100, True)
blobs.filter_blobs(10, 10000)
blob_count = blobs.GetNumBlobs()
# Find blob closest to last ball position
if ball_last:
last_center = blob_center(ball_last)
mind2 = 1e100
ball = None
for i in range(blob_count):
blob = blobs.GetBlob(i)
center = blob_center(blob)
d2 = distance_sq(last_center, center)
if d2 < mind2:
mind2 = d2
ball = blob
if ball:
cvRectangle(frame, (ball.MinX(), ball.MinY()),
(ball.MaxX(), ball.MaxY()), CV_RGB(0, 255, 0))
ball_last = ball
# Initialize ball estimate
else:
possible_ball = []
for i in range(blob_count):
blob = blobs.GetBlob(i)
x, y = blob_center(blob)
if 40 < y < 142:
possible_ball.append(blob)
# Find fastest-moving possible ball
maxv = 3
ball = None
for blob in possible_ball:
#blob.FillBlob(frame, cvScalar(0, 255, 0), 0, 0)
x1, y1 = int(blob.MinX()), int(blob.MinY())
x2, y2 = int(blob.MaxX()), int(blob.MaxY())
# get mean optical flow velocity over bounding box
vx = numpy.mean(velx[y1:y2, x1:x2])
vy = numpy.mean(vely[y1:y2, x1:x2])
'''
# get mean at center of blob
x, y = int((x1+x2)/2), int((y1+y2)/2)
vx = velx[y,x]
vy = vely[y,x]
'''
# v = proportional to sq magnitude of velocity vector
v = vx * vx + vy * vy
print v
if v > maxv:
ball = blob
maxv = v
if ball:
cvRectangle(frame, (ball.MinX(), ball.MinY()),
(ball.MaxX(), ball.MaxY()),
CV_RGB(0, 255, 0))
ball_last = ball
cvShowImage('video', frame)
'''
# blobs
myblobs = CBlobResult(edges, blob_mask, 100, True)
myblobs.filter_blobs(10, 10000)
blob_count = myblobs.GetNumBlobs()
near_player = []
far_player = []
possible_ball = []
for i in range(blob_count):
blob = myblobs.GetBlob(i)
# Group blobs by their center point
x, y = blob_center(blob)
if y > 142:
near_player.append(blob)
elif y < 40:
far_player.append(blob)
else:
possible_ball.append(blob)
outliers = [remove_outlier(near_player),
remove_outlier(far_player)]
outliers = [x for x in outliers if x]
print outliers
# If no ball found in the range y=(40, 145) then
# find the closest blob to our state estimate
# that is small enough to be a ball.
if len(possible_ball) == 0:
outliers = [blob for blob in outliers if blob.Area() < 100]
if len(outliers) > 0:
center = cvKalmanPredict(kf_ball)
outliers.sort(key=lambda x: \
distance_sq(blob_center(x), center))
possible_ball.append(outliers[0])
ball = None
maxv = -1
# Find the ball that's moving the fastest
for blob in possible_ball:
# get optical flow velocity at blob center
x, y = blob_center(blob)
x, y = int(x), int(y)
# v = magnitude of velocity vector
v = pow(cvGetReal2D(velx, y, x), 2) + \
pow(cvGetReal2D(vely, y, x), 2)
if v > maxv:
ball = blob
maxv = v
if ball and maxv > 0.01:
cvRectangle(frame, (ball.MinX(), ball.MinY()),
(ball.MaxX(), ball.MaxY()), CV_RGB(0, 255, 0))
#ball.FillBlob(frame, cvScalar(0, 255, 0), 0, 0)
# Estimate position of near player
if len(near_player) > 0:
# Find x center
xLocs = []
yLocs = []
for blob in near_player:
x, y = blob_center(blob)
xLocs.append(x)
yLocs.append(y)
if len(xLocs) > 0:
xmin, ymin = min(xLocs), min(yLocs)
xmax, ymax = max(xLocs), max(yLocs)
cvRectangle(frame, (xmin, ymin), (xmax, ymax),
CV_RGB(255, 0, 0))
y_k = cvKalmanPredict(kf_near)
print 'near player at', y_k[0], y_k[1]
draw_cross(frame, y_k, CV_RGB(255, 0, 0), 10)
# z1, z2 = position measurement
z_k[0], z_k[1] = (xmin+xmax)/2, (ymin+ymax)/2
cvMatMulAdd(kf_near.measurement_matrix, x_near, z_k, z_k)
cvKalmanCorrect(kf_near, z_k)
# apply process noise and update state
cvRandArr(rng, w_k, CV_RAND_NORMAL, 0,
numpy.sqrt(kf_near.process_noise_cov[0,0]))
cvMatMulAdd(kf_near.transition_matrix, x_near, w_k, x_near)
# Estimate position of far player
xLocs = []
yLocs = []
for blob in far_player:
x,y = blob_center(blob)
xLocs.append(x)
yLocs.append(y)
if len(xLocs) > 0:
xmin, ymin = min(xLocs), min(yLocs)
xmax, ymax = max(xLocs), max(yLocs)
cvRectangle(frame, (xmin, ymin), (xmax, ymax),
CV_RGB(0, 0, 255))
print 'far player at', (xmin+xmax)/2, (ymin+ymax)/2
# Estimate position of the ball
y_k = cvKalmanPredict(kf_ball)
x, y = y_k[0], y_k[1]
print 'ball at', x, y
draw_cross(frame, y_k, CV_RGB(0, 255, 0), 5)
# kalman filter update
if ball and maxv > 0.01:
# z1, z2 = position measurement
x, y = blob_center(ball)
z_k[0], z_k[1] = x, y
# z3, z4 = velocity measurement
#x, y = int(x), int(y)
#z_k[2], z_k[3] = cvGetReal2D(velx, y, x), \
# cvGetReal2D(vely, y, x)
cvMatMulAdd(kf_ball.measurement_matrix, x_ball, z_k, z_k)
cvKalmanCorrect(kf_ball, z_k)
# apply process noise and update state
cvRandArr(rng, w_k, CV_RAND_NORMAL, 0,
numpy.sqrt(kf_ball.process_noise_cov[0,0]))
cvMatMulAdd(kf_ball.transition_matrix, x_ball, w_k, x_ball)
#else:
#kf_ball.transition_matrix[0,2] = 0
#kf_ball.transition_matrix[1,3] = 0
#x_ball = cvKalmanPredict(kf_ball)
'''
''' crashes
#hough transform on dilated image
cvCopy(edges, hough_in)
cvSmooth(hough_in, hough_in, CV_GAUSSIAN, 15, 15, 0, 0)
cvHoughCircles(hough_in, hough_storage, CV_HOUGH_GRADIENT,
4, frame_size[1]/10, 100, 40, 0, 0)
print hough_storage
'''
k = cvWaitKey(1000 / fps)
if k == 27: # ESC key
break
elif k == 'p': # play/pause
play = not play
def main():
if len(sys.argv) == 1:
print 'Usage: %s [inputfile]' % sys.argv[0]
sys.exit(1)
# initialize window
cvNamedWindow('video', CV_WINDOW_AUTOSIZE)
cvMoveWindow('video', 10, 10)
cvNamedWindow('segment', CV_WINDOW_AUTOSIZE)
cvMoveWindow('segment', 10, 500)
cvNamedWindow('flow', CV_WINDOW_AUTOSIZE)
cvMoveWindow('flow', 500, 10)
cvNamedWindow('edges', CV_WINDOW_AUTOSIZE)
cvMoveWindow('edges', 500, 500)
capture = cvCreateFileCapture(sys.argv[1])
if not capture: print 'Error opening capture'; sys.exit(1)
# Load bg image
bg = cvLoadImage('bg.png')
# Discard some frames
for i in xrange(2300):
cvGrabFrame(capture)
frame = cvQueryFrame(capture)
frame_size = cvGetSize(frame)
# vars for playback
fps = 25
play = True
velx = cvCreateImage(frame_size, IPL_DEPTH_32F, 1)
vely = cvCreateImage(frame_size, IPL_DEPTH_32F, 1)
combined = cvCreateImage(frame_size, IPL_DEPTH_8U, 1)
prev = cvCreateImage(frame_size, IPL_DEPTH_8U, 1)
frame_gray = cvCreateImage(frame_size, IPL_DEPTH_8U, 1)
frame_sub = cvCreateImage(frame_size, IPL_DEPTH_8U, 3)
flow = cvCreateImage(frame_size, IPL_DEPTH_8U, 3)
edges = cvCreateImage(frame_size, IPL_DEPTH_8U, 1)
prev_edges = cvCreateImage(frame_size, IPL_DEPTH_8U, 1)
storage = cvCreateMemStorage(0)
blob_mask = cvCreateImage (frame_size, IPL_DEPTH_8U, 1)
cvSet(blob_mask, 1)
hough_in = cvCreateImage(frame_size, IPL_DEPTH_8U, 1)
hough_storage = cvCreateMat( 100, 1, CV_32FC3 )
silhouette = cvCreateImage(frame_size, IPL_DEPTH_8U, 1)
mhi = cvCreateImage(frame_size, IPL_DEPTH_32F, 1)
seg_mask = cvCreateImage(frame_size, IPL_DEPTH_32F, 1)
seg_storage = cvCreateMemStorage(0)
kf_near = cvCreateKalman(4, 2, 0)
kf_far = cvCreateKalman(4, 2, 0)
kf_ball = cvCreateKalman(4, 2, 0)
for kalman in [kf_near, kf_far, kf_ball]:
# transition matrix F is initialized to identity
# want F = [1 0 1 0
# 0 1 0 1
# 0 0 1 0
# 0 0 0 1]
# because x1 = x coordinate and x3 = d/dt x1
# x2 = y coordinate and x4 = d/dt x3
kalman.transition_matrix[0,2] = 1
kalman.transition_matrix[1,3] = 1
cvSetIdentity(kalman.measurement_matrix, 1)
cvSetIdentity(kalman.process_noise_cov, 1e-5)
cvSetIdentity(kalman.measurement_noise_cov, 1e-1)
cvSetIdentity(kalman.error_cov_post, 1)
cvSetIdentity(kf_ball.process_noise_cov, 1e-3)
# initialize random number generator
rng = cvRNG()
# x_k = state variables
x_ball = cvCreateMat(4, 1, CV_32FC1)
cvZero(x_ball)
cvZero(kf_ball.state_post)
x_near = cvCreateMat(4, 1, CV_32FC1)
cvZero(x_near)
cvZero(kf_near.state_post)
# w_k = noise
w_k = cvCreateMat(4, 1, CV_32FC1)
# z_k = measurements
z_k = cvCreateMat(2, 1, CV_32FC1)
ball_last = None
while True:
if play:
frame = cvQueryFrame(capture)
# do background subtraction using manually created frame
# TODO do averaging to generate bg
cvSub(frame, bg, frame_sub)
# convert to grayscale
cvCvtColor(frame_sub, frame_gray, CV_BGR2GRAY)
# edge detection/dilation
cvCanny(frame_gray, edges, 50, 100)
cvDilate(edges, edges,
cvCreateStructuringElementEx(7, 7, 0, 0,
CV_SHAPE_ELLIPSE))
#cvErode(edges, edges)
cvShowImage('edges', edges)
# do optical flow using edge detection/dilation image
#cvCalcOpticalFlowLK(prev_edges, edges, (7,7), velx, vely)
cvCalcOpticalFlowHS(prev_edges, edges, 1, velx, vely, 0.5,
(CV_TERMCRIT_ITER, 10, 0))
#cvThreshold(vely, vely, 0.5, 0, CV_THRESH_TOZERO)
#cvAdd(velx, vely, vely)
cvShowImage('flow', vely)
cvCopy(edges, prev_edges)
'''
# segment motion
cvThreshold(vely, silhouette, 0.5, 255, CV_THRESH_BINARY)
cvUpdateMotionHistory(silhouette, mhi, time.clock()/1000, 1)
components = cvSegmentMotion(mhi, seg_mask, seg_storage,
time.clock()/1000,
1/fps)
#http://opencv.willowgarage.com/documentation/python/imgproc_miscellaneous_image_transformations.html?highlight=connectedcomp#CvConnectedComp
for c in components:
area, value, rect = c.area, c.value, c.rect
x, y, w, h = rect.x, rect.y, rect.width, rect.height
cvRectangle(frame, (x,y), (x+w,y+h), CV_RGB(255, 0, 0))
cvShowImage('segment', seg_mask)
'''
blobs = CBlobResult(edges, blob_mask, 100, True)
blobs.filter_blobs(10, 10000)
blob_count = blobs.GetNumBlobs()
# Find blob closest to last ball position
if ball_last:
last_center = blob_center(ball_last)
mind2 = 1e100
ball = None
for i in range(blob_count):
blob = blobs.GetBlob(i)
center = blob_center(blob)
d2 = distance_sq(last_center, center)
if d2 < mind2:
mind2 = d2
ball = blob
if ball:
cvRectangle(frame, (ball.MinX(), ball.MinY()),
(ball.MaxX(), ball.MaxY()),
CV_RGB(0, 255, 0))
ball_last = ball
# Initialize ball estimate
else:
possible_ball = []
for i in range(blob_count):
blob = blobs.GetBlob(i)
x, y = blob_center(blob)
if 40 < y < 142:
possible_ball.append(blob)
# Find fastest-moving possible ball
maxv = 3
ball = None
for blob in possible_ball:
#blob.FillBlob(frame, cvScalar(0, 255, 0), 0, 0)
x1, y1 = int(blob.MinX()), int(blob.MinY())
x2, y2 = int(blob.MaxX()), int(blob.MaxY())
# get mean optical flow velocity over bounding box
vx = numpy.mean(velx[y1:y2, x1:x2])
vy = numpy.mean(vely[y1:y2, x1:x2])
'''
# get mean at center of blob
x, y = int((x1+x2)/2), int((y1+y2)/2)
vx = velx[y,x]
vy = vely[y,x]
'''
# v = proportional to sq magnitude of velocity vector
v = vx*vx + vy*vy
print v
if v > maxv:
ball = blob
maxv = v
if ball:
cvRectangle(frame, (ball.MinX(), ball.MinY()),
(ball.MaxX(), ball.MaxY()),
CV_RGB(0, 255, 0))
ball_last = ball
cvShowImage('video', frame)
'''
# blobs
myblobs = CBlobResult(edges, blob_mask, 100, True)
myblobs.filter_blobs(10, 10000)
blob_count = myblobs.GetNumBlobs()
near_player = []
far_player = []
possible_ball = []
for i in range(blob_count):
blob = myblobs.GetBlob(i)
# Group blobs by their center point
x, y = blob_center(blob)
if y > 142:
near_player.append(blob)
elif y < 40:
far_player.append(blob)
else:
possible_ball.append(blob)
outliers = [remove_outlier(near_player),
remove_outlier(far_player)]
outliers = [x for x in outliers if x]
print outliers
# If no ball found in the range y=(40, 145) then
# find the closest blob to our state estimate
# that is small enough to be a ball.
if len(possible_ball) == 0:
outliers = [blob for blob in outliers if blob.Area() < 100]
if len(outliers) > 0:
center = cvKalmanPredict(kf_ball)
outliers.sort(key=lambda x: \
distance_sq(blob_center(x), center))
possible_ball.append(outliers[0])
ball = None
maxv = -1
# Find the ball that's moving the fastest
for blob in possible_ball:
# get optical flow velocity at blob center
x, y = blob_center(blob)
x, y = int(x), int(y)
# v = magnitude of velocity vector
v = pow(cvGetReal2D(velx, y, x), 2) + \
pow(cvGetReal2D(vely, y, x), 2)
if v > maxv:
ball = blob
maxv = v
if ball and maxv > 0.01:
cvRectangle(frame, (ball.MinX(), ball.MinY()),
(ball.MaxX(), ball.MaxY()), CV_RGB(0, 255, 0))
#ball.FillBlob(frame, cvScalar(0, 255, 0), 0, 0)
# Estimate position of near player
if len(near_player) > 0:
# Find x center
xLocs = []
yLocs = []
for blob in near_player:
x, y = blob_center(blob)
xLocs.append(x)
yLocs.append(y)
if len(xLocs) > 0:
xmin, ymin = min(xLocs), min(yLocs)
xmax, ymax = max(xLocs), max(yLocs)
cvRectangle(frame, (xmin, ymin), (xmax, ymax),
CV_RGB(255, 0, 0))
y_k = cvKalmanPredict(kf_near)
print 'near player at', y_k[0], y_k[1]
draw_cross(frame, y_k, CV_RGB(255, 0, 0), 10)
# z1, z2 = position measurement
z_k[0], z_k[1] = (xmin+xmax)/2, (ymin+ymax)/2
cvMatMulAdd(kf_near.measurement_matrix, x_near, z_k, z_k)
cvKalmanCorrect(kf_near, z_k)
# apply process noise and update state
cvRandArr(rng, w_k, CV_RAND_NORMAL, 0,
numpy.sqrt(kf_near.process_noise_cov[0,0]))
cvMatMulAdd(kf_near.transition_matrix, x_near, w_k, x_near)
# Estimate position of far player
xLocs = []
yLocs = []
for blob in far_player:
x,y = blob_center(blob)
xLocs.append(x)
yLocs.append(y)
if len(xLocs) > 0:
xmin, ymin = min(xLocs), min(yLocs)
xmax, ymax = max(xLocs), max(yLocs)
cvRectangle(frame, (xmin, ymin), (xmax, ymax),
CV_RGB(0, 0, 255))
print 'far player at', (xmin+xmax)/2, (ymin+ymax)/2
# Estimate position of the ball
y_k = cvKalmanPredict(kf_ball)
x, y = y_k[0], y_k[1]
print 'ball at', x, y
draw_cross(frame, y_k, CV_RGB(0, 255, 0), 5)
# kalman filter update
if ball and maxv > 0.01:
# z1, z2 = position measurement
x, y = blob_center(ball)
z_k[0], z_k[1] = x, y
# z3, z4 = velocity measurement
#x, y = int(x), int(y)
#z_k[2], z_k[3] = cvGetReal2D(velx, y, x), \
# cvGetReal2D(vely, y, x)
cvMatMulAdd(kf_ball.measurement_matrix, x_ball, z_k, z_k)
cvKalmanCorrect(kf_ball, z_k)
# apply process noise and update state
cvRandArr(rng, w_k, CV_RAND_NORMAL, 0,
numpy.sqrt(kf_ball.process_noise_cov[0,0]))
cvMatMulAdd(kf_ball.transition_matrix, x_ball, w_k, x_ball)
#else:
#kf_ball.transition_matrix[0,2] = 0
#kf_ball.transition_matrix[1,3] = 0
#x_ball = cvKalmanPredict(kf_ball)
'''
''' crashes
#hough transform on dilated image
cvCopy(edges, hough_in)
cvSmooth(hough_in, hough_in, CV_GAUSSIAN, 15, 15, 0, 0)
cvHoughCircles(hough_in, hough_storage, CV_HOUGH_GRADIENT,
4, frame_size[1]/10, 100, 40, 0, 0)
print hough_storage
'''
k = cvWaitKey(1000/fps)
if k == 27: # ESC key
break
elif k == 'p': # play/pause
play = not play
def main():
if len(sys.argv) == 1:
print 'Usage: %s [inputfile]' % sys.argv[0]
sys.exit(1)
# initialize window
cvNamedWindow('video', CV_WINDOW_AUTOSIZE)
cvMoveWindow('video', 10, 10)
cvNamedWindow('threshold', CV_WINDOW_AUTOSIZE)
cvMoveWindow('threshold', 10, 500)
cvNamedWindow('flow', CV_WINDOW_AUTOSIZE)
cvMoveWindow('flow', 500, 10)
cvNamedWindow('edges', CV_WINDOW_AUTOSIZE)
cvMoveWindow('edges', 500, 500)
cvNamedWindow('combined', CV_WINDOW_AUTOSIZE)
cvMoveWindow('combined', 10, 1000)
capture = cvCreateFileCapture(sys.argv[1])
if not capture: print 'Error opening capture'; sys.exit(1)
# Load bg image
bg = cvLoadImage('bg.png')
# Discard some frames
for i in xrange(2300):
cvGrabFrame(capture)
frame = cvQueryFrame(capture)
frame_size = cvGetSize(frame)
# vars for playback
fps = 25
play = True
velx = cvCreateImage(frame_size, IPL_DEPTH_32F, 1)
vely = cvCreateImage(frame_size, IPL_DEPTH_32F, 1)
combined = cvCreateImage(frame_size, IPL_DEPTH_8U, 1)
prev = cvCreateImage(frame_size, IPL_DEPTH_8U, 1)
curr = cvCreateImage(frame_size, IPL_DEPTH_8U, 1)
frame_sub = cvCreateImage(frame_size, IPL_DEPTH_8U, 3)
flow = cvCreateImage(frame_size, IPL_DEPTH_8U, 3)
edges = cvCreateImage(frame_size, IPL_DEPTH_8U, 1)
prev_edges = None
storage = cvCreateMemStorage(0)
blob_mask = cvCreateImage (frame_size, IPL_DEPTH_8U, 1)
cvSet(blob_mask, 1)
hough_in = cvCreateImage(frame_size, IPL_DEPTH_8U, 1)
hough_storage = cvCreateMat( 100, 1, CV_32FC3 )
seg_mask = cvCreateImage(frame_size, IPL_DEPTH_32F, 1)
seg_storage = cvCreateMemStorage(0)
'''
cvCvtColor(frame, prev, CV_BGR2GRAY)
frame = cvQueryFrame(capture)
cvCvtColor(frame, curr, CV_BGR2GRAY)
# winSize can't have even numbers
cvCalcOpticalFlowLK(prev, curr, (3,3), velx, vely)
cvShowImage('video', frame)
cvShowImage('flow', velx)
cvWaitKey(0)
'''
kf_near = cvCreateKalman(4, 2, 0)
kf_far = cvCreateKalman(4, 2, 0)
kf_ball = cvCreateKalman(4, 2, 0)
for kalman in [kf_near, kf_far, kf_ball]:
# transition matrix F is initialized to identity
# want F = [1 0 1 0
# 0 1 0 1
# 0 0 1 0
# 0 0 0 1]
# because x1 = x coordinate and x3 = d/dt x1
# x2 = y coordinate and x4 = d/dt x3
kalman.transition_matrix[0,2] = 1
kalman.transition_matrix[1,3] = 1
cvSetIdentity(kalman.measurement_matrix, 1)
cvSetIdentity(kalman.process_noise_cov, 1e-5)
cvSetIdentity(kalman.measurement_noise_cov, 1e-1)
cvSetIdentity(kalman.error_cov_post, 1)
cvSetIdentity(kf_ball.process_noise_cov, 1e-3)
# initialize random number generator
rng = cvRNG()
# x_k = state variables
x_ball = cvCreateMat(4, 1, CV_32FC1)
cvZero(x_ball)
cvZero(kf_ball.state_post)
x_near = cvCreateMat(4, 1, CV_32FC1)
cvZero(x_near)
cvZero(kf_near.state_post)
# w_k = noise
w_k = cvCreateMat(4, 1, CV_32FC1)
# z_k = measurements
z_k = cvCreateMat(2, 1, CV_32FC1)
while True:
if play:
frame = cvQueryFrame(capture)
cvSub(frame, bg, frame_sub)
'''#detect people
found = list(cvHOGDetectMultiScale(frame, storage, win_stride=(8,8),
padding=(32,32), scale=1.05, group_threshold=2))
for r in found:
(rx, ry), (rw, rh) = r
tl = (rx + int(rw*0.1), ry + int(rh*0.07))
br = (rx + int(rw*0.9), ry + int(rh*0.87))
cvRectangle(frame, tl, br, (0, 255, 0), 3)
'''
#color thresholding
hsv = cvCreateImage(frame_size, IPL_DEPTH_8U, 3);
cvCvtColor(frame, hsv, CV_BGR2HSV)
mask = cvCreateMat(frame_size.height, frame_size.width,
CV_8UC1)
cvInRangeS(hsv, (0.03*256, 0.2*256, 0.6*256, 0),
(0.16*256, 1.0*256, 1.0*256, 0), mask)
cvShowImage('threshold', mask)
#optical flow method
# store previous frame
prev, curr = curr, prev
# convert next frame to single channel grayscale
cvCvtColor(frame_sub, curr, CV_BGR2GRAY)
#cvCalcOpticalFlowLK(prev, curr, (3,3), velx, vely)
#cvThreshold(velx, velx, 8.0, 0, CV_THRESH_TOZERO)
cvCalcOpticalFlowHS(prev, curr, 1, velx, vely, 0.5,
(CV_TERMCRIT_ITER, 10, 0))
cvThreshold(velx, velx, 0.5, 0, CV_THRESH_TOZERO)
cvThreshold(vely, vely, 0.5, 0, CV_THRESH_TOZERO)
cvErode(vely, vely,
cvCreateStructuringElementEx(2, 2, 0, 0,
CV_SHAPE_ELLIPSE))
cvAdd(vely, velx, vely)
cvShowImage('flow', vely)
#edge detection/dilation
cvCanny(curr, edges, 50, 100)
cvDilate(edges, edges,
cvCreateStructuringElementEx(7, 7, 0, 0,
CV_SHAPE_ELLIPSE))
#cvErode(edges, edges)
cvShowImage('edges', edges)
# do optical flow using edge detection/dilation
if prev_edges:
cvCalcOpticalFlowHS(prev_edges, edges, 1, velx, vely, 0.5,
(CV_TERMCRIT_ITER, 10, 0))
#cvThreshold(vely, vely, 0.5, 0, CV_THRESH_TOZERO)
cvShowImage('flow', vely)
prev_edges = edges
cvThreshold(vely, combined, 0.5, 255, CV_THRESH_BINARY)
cvMin(combined, edges, combined)
cvShowImage('combined', combined)
# blobs
myblobs = CBlobResult(edges, blob_mask, 100, True)
myblobs.filter_blobs(10, 10000)
blob_count = myblobs.GetNumBlobs()
near_player = []
far_player = []
possible_ball = []
for i in range(blob_count):
blob = myblobs.GetBlob(i)
# Group blobs by their center point
x, y = blob_center(blob)
if y > 142:
near_player.append(blob)
elif y < 40:
far_player.append(blob)
else:
possible_ball.append(blob)
'''
# Remove outliers - this was commented out because
# it's not as effective as hoped
outliers = [remove_outlier(near_player),
remove_outlier(far_player)]
outliers = [x for x in outliers if x]
print outliers
# If no ball found in the range y=(40, 145) then
# find the closest blob to our state estimate
# that is small enough to be a ball.
if len(possible_ball) == 0:
outliers = [blob for blob in outliers if blob.Area() < 100]
if len(outliers) > 0:
center = cvKalmanPredict(kf_ball)
outliers.sort(key=lambda x: \
distance_sq(blob_center(x), center))
possible_ball.append(outliers[0])
'''
ball = None
maxv = -1
# Find the ball that's moving the fastest
for blob in possible_ball:
# get optical flow velocity at blob center
x, y = blob_center(blob)
x, y = int(x), int(y)
# v = magnitude of velocity vector
v = pow(cvGetReal2D(velx, y, x), 2) + \
pow(cvGetReal2D(vely, y, x), 2)
if v > maxv:
ball = blob
maxv = v
if ball and maxv > 0.01:
x, y = blob_center(ball)
#cvRectangle(frame, (ball.MinX(), ball.MinY()),
# (ball.MaxX(), ball.MaxY()), CV_RGB(0, 255, 0))
cvRectangle(frame, (x-5,y-5), (x+5,y+5), CV_RGB(0, 255, 0))
#ball.FillBlob(frame, cvScalar(0, 255, 0), 0, 0)
# Estimate position of near player
if len(near_player) > 0:
# Find x center
xLocs = []
yLocs = []
for blob in near_player:
x, y = blob_center(blob)
xLocs.append(x)
yLocs.append(y)
if len(xLocs) > 0:
xmin, ymin = min(xLocs), min(yLocs)
xmax, ymax = max(xLocs), max(yLocs)
#cvRectangle(frame, (xmin, ymin), (xmax, ymax),
# CV_RGB(255, 0, 0))
y_k = cvKalmanPredict(kf_near)
print 'near player at', y_k[0], y_k[1]
#draw_cross(frame, y_k, CV_RGB(255, 0, 0), 10)
# z1, z2 = position measurement
z_k[0], z_k[1] = (xmin+xmax)/2, (ymin+ymax)/2
cvMatMulAdd(kf_near.measurement_matrix, x_near, z_k, z_k)
cvKalmanCorrect(kf_near, z_k)
# apply process noise and update state
cvRandArr(rng, w_k, CV_RAND_NORMAL, 0,
numpy.sqrt(kf_near.process_noise_cov[0,0]))
cvMatMulAdd(kf_near.transition_matrix, x_near, w_k, x_near)
# Estimate position of far player
xLocs = []
yLocs = []
for blob in far_player:
x,y = blob_center(blob)
xLocs.append(x)
yLocs.append(y)
if len(xLocs) > 0:
xmin, ymin = min(xLocs), min(yLocs)
xmax, ymax = max(xLocs), max(yLocs)
#cvRectangle(frame, (xmin, ymin), (xmax, ymax),
# CV_RGB(0, 0, 255))
print 'far player at', (xmin+xmax)/2, (ymin+ymax)/2
# Estimate position of the ball
y_k = cvKalmanPredict(kf_ball)
x, y = y_k[0], y_k[1]
print 'ball at', x, y
#draw_cross(frame, y_k, CV_RGB(0, 255, 0), 5)
# kalman filter update
if ball and maxv > 0.01:
# z1, z2 = position measurement
x, y = blob_center(ball)
z_k[0], z_k[1] = x, y
# z3, z4 = velocity measurement
#x, y = int(x), int(y)
#z_k[2], z_k[3] = cvGetReal2D(velx, y, x), \
# cvGetReal2D(vely, y, x)
cvMatMulAdd(kf_ball.measurement_matrix, x_ball, z_k, z_k)
cvKalmanCorrect(kf_ball, z_k)
# apply process noise and update state
cvRandArr(rng, w_k, CV_RAND_NORMAL, 0,
numpy.sqrt(kf_ball.process_noise_cov[0,0]))
cvMatMulAdd(kf_ball.transition_matrix, x_ball, w_k, x_ball)
#else:
#kf_ball.transition_matrix[0,2] = 0
#kf_ball.transition_matrix[1,3] = 0
#x_ball = cvKalmanPredict(kf_ball)
cvShowImage('video', frame)
''' crashes
#hough transform on dilated image
cvCopy(edges, hough_in)
cvSmooth(hough_in, hough_in, CV_GAUSSIAN, 15, 15, 0, 0)
cvHoughCircles(hough_in, hough_storage, CV_HOUGH_GRADIENT,
4, frame_size[1]/10, 100, 40, 0, 0)
print hough_storage
'''
k = cvWaitKey(1000/fps)
if k == 27: # ESC key
break
elif k == 'p': # play/pause
play = not play
def main():
if len(sys.argv) == 1:
print 'Usage: %s [inputfile]' % sys.argv[0]
sys.exit(1)
# initialize window
cv.NamedWindow('video', cv.CV_WINDOW_AUTOSIZE)
cv.MoveWindow('video', 10, 10)
cv.NamedWindow('threshold', cv.CV_WINDOW_AUTOSIZE)
cv.MoveWindow('threshold', 10, 500)
cv.NamedWindow('flow', cv.CV_WINDOW_AUTOSIZE)
cv.MoveWindow('flow', 500, 10)
cv.NamedWindow('edges', cv.CV_WINDOW_AUTOSIZE)
cv.MoveWindow('edges', 500, 500)
cv.NamedWindow('combined', cv.CV_WINDOW_AUTOSIZE)
cv.MoveWindow('combined', 1000, 10)
capture = cv.CreateFileCapture(sys.argv[1])
if not capture:
print 'Error opening capture'
sys.exit(1)
# Load bg image
bg = cv.LoadImage('bg.png')
# Discard some frames
for i in xrange(2300):
cv.GrabFrame(capture)
frame = cv.QueryFrame(capture)
frame_size = cv.GetSize(frame)
# vars for playback
fps = 25
play = True
velx = cv.CreateImage(frame_size, cv.IPL_DEPTH_32F, 1)
vely = cv.CreateImage(frame_size, cv.IPL_DEPTH_32F, 1)
combined = cv.CreateImage(frame_size, cv.IPL_DEPTH_8U, 1)
prev = cv.CreateImage(frame_size, cv.IPL_DEPTH_8U, 1)
curr = cv.CreateImage(frame_size, cv.IPL_DEPTH_8U, 1)
frame_sub = cv.CreateImage(frame_size, cv.IPL_DEPTH_8U, 3)
edges = cv.CreateImage(frame_size, cv.IPL_DEPTH_8U, 1)
prev_edges = None
storage = cv.CreateMemStorage(0)
blob_mask = cv0.cvCreateImage(frame_size, cv.IPL_DEPTH_8U, 1)
cv0.cvSet(blob_mask, 1)
hough_in = cv.CreateImage(frame_size, cv.IPL_DEPTH_8U, 1)
hough_storage = cv.CreateMat(100, 1, cv.CV_32FC3)
'''
cv.CvtColor(frame, prev, cv.CV_BGR2GRAY)
frame = cv.QueryFrame(capture)
cv.CvtColor(frame, curr, cv.CV_BGR2GRAY)
# winSize can't have even numbers
cv.CalcOpticalFlowLK(prev, curr, (3,3), velx, vely)
cv.ShowImage('video', frame)
cv.ShowImage('flow', velx)
cv.WaitKey(0)
'''
while True:
if play:
frame = cv.QueryFrame(capture)
cv.Sub(frame, bg, frame_sub)
'''#detect people
found = list(cv.HOGDetectMultiScale(frame, storage, win_stride=(8,8),
padding=(32,32), scale=1.05, group_threshold=2))
for r in found:
(rx, ry), (rw, rh) = r
tl = (rx + int(rw*0.1), ry + int(rh*0.07))
br = (rx + int(rw*0.9), ry + int(rh*0.87))
cv.Rectangle(frame, tl, br, (0, 255, 0), 3)
'''
#color thresholding
hsv = cv.CreateImage(frame_size, cv.IPL_DEPTH_8U, 3)
cv.CvtColor(frame, hsv, cv.CV_BGR2HSV)
mask = cv.CreateMat(frame_size[1], frame_size[0], cv.CV_8UC1)
cv.InRangeS(hsv, (0.06 * 256, 0.2 * 256, 0.6 * 256, 0),
(0.16 * 256, 1.0 * 256, 1.0 * 256, 0), mask)
cv.ShowImage('threshold', mask)
#optical flow method
# store previous frame
prev, curr = curr, prev
# convert next frame to single channel grayscale
cv.CvtColor(frame_sub, curr, cv.CV_BGR2GRAY)
#cv.CalcOpticalFlowLK(prev, curr, (3,3), velx, vely)
#cv.Threshold(velx, velx, 8.0, 0, cv.CV_THRESH_TOZERO)
cv.CalcOpticalFlowHS(prev, curr, 1, velx, vely, 0.5,
(cv.CV_TERMCRIT_ITER, 10, 0))
cv.Threshold(velx, velx, 0.5, 0, cv.CV_THRESH_TOZERO)
cv.Threshold(vely, vely, 0.5, 0, cv.CV_THRESH_TOZERO)
cv.Erode(
vely, vely,
cv.CreateStructuringElementEx(2, 2, 0, 0, cv.CV_SHAPE_ELLIPSE))
cv.Add(vely, velx, vely)
cv.ShowImage('flow', vely)
#edge detection
cv.Canny(curr, edges, 50, 100)
cv.Dilate(
edges, edges,
cv.CreateStructuringElementEx(7, 7, 0, 0, cv.CV_SHAPE_ELLIPSE))
cv.ShowImage('edges', edges)
if prev_edges:
cv.CalcOpticalFlowHS(prev_edges, edges, 1, velx, vely, 0.5,
(cv.CV_TERMCRIT_ITER, 10, 0))
cv.Threshold(velx, velx, 0.5, 0, cv.CV_THRESH_TOZERO)
cv.Threshold(vely, vely, 0.5, 0, cv.CV_THRESH_TOZERO)
cv.ShowImage('flow', vely)
prev_edges = edges
cv.Threshold(vely, combined, 0.5, 255, cv.CV_THRESH_BINARY)
cv.Min(combined, edges, combined)
cv.ShowImage('combined', combined)
# blobs
myblobs = CBlobResult(edges, blob_mask, 100, False)
myblobs.filter_blobs(10, 10000)
blob_count = myblobs.GetNumBlobs()
for i in range(blob_count):
my_enumerated_blob = myblobs.GetBlob(i)
# print "%d: Area = %d" % (i, my_enumerated_blob.Area())
my_enumerated_blob.FillBlob(frame,
hsv2rgb(i * 180.0 / blob_count), 0,
0)
cv.ShowImage('video', frame)
''' crashes
#hough transform on dilated image
#http://wiki.elphel.com/index.php?
# title=OpenCV_Tennis_balls_recognizing_tutorial&redirect=no
cv.Copy(edges, hough_in)
cv.Smooth(hough_in, hough_in, cv.CV_GAUSSIAN, 15, 15, 0, 0)
cv.HoughCircles(hough_in, hough_storage, cv.CV_HOUGH_GRADIENT,
4, frame_size[1]/10, 100, 40, 0, 0)
print hough_storage
'''
k = cv.WaitKey(1000 / fps)
if k == 27: # ESC key
break
elif k == 'p': # play/pause
play = not play
# 1. capture the current image
frame = highgui.cvQueryFrame(capture)
if frame is None:
# no image captured... end the processing
break
# mirror the captured image
cv.cvFlip(frame, None, 1)
cv.cvCvtColor(frame, my_grayscale, cv.CV_RGB2GRAY)
cv.cvThreshold(my_grayscale, my_grayscale, 128, 255, cv.CV_THRESH_BINARY)
if not blob_overlay:
# Convert black-and-white version back into three-color representation
cv.cvCvtColor(my_grayscale, frame, cv.CV_GRAY2RGB)
myblobs = CBlobResult(my_grayscale, mask, 100, True)
myblobs.filter_blobs(10, 10000)
blob_count = myblobs.GetNumBlobs()
for i in range(blob_count):
my_enumerated_blob = myblobs.GetBlob(i)
# print "%d: Area = %d" % (i, my_enumerated_blob.Area())
my_enumerated_blob.FillBlob(frame, hsv2rgb(i * 180.0 / blob_count), 0, 0)
# we can now display the images
highgui.cvShowImage("Blob View", frame)
# handle events
k = highgui.cvWaitKey(10)
