def track(self, pathid, start, stop, basepath, paths):
if pathid not in paths:
return Path(None, None, {})
path = paths[pathid]
if start not in path.boxes:
return Path(path.label, path.id, {})
startbox = path.boxes[start]
initialrect = [startbox.xtl, startbox.ytl, startbox.xbr-startbox.xtl, startbox.ybr-startbox.ytl]
startbox = [startbox.xtl, startbox.ytl, startbox.xbr, startbox.ybr]
frames = getframes(basepath, False)
previmage = frames[start]
imagesize = previmage.shape
#print "Frame shape: "
#print previmage.shape
#print '=================='
#if not self.started:
self.tracker.start_track(frames[start], startbox)#self.visbox_to_dbox(startbox))
self.started = True
boxes = self.dlib_track(start, stop, frames, initialrect, imagesize)
#meanshift(start, stop, initialrect, imagesize)
# boxes need to be in vision.Box() form
# width is computed from x's, so corresponds to the columns, y to the rows (see documentation in annotation.py
#[dbox.left(), dbox.top(), dbox.right(), dbox.bottom()]
return Path(path.label, path.id, boxes)
def bidirectional(tracker, start, stop, basepath, pathid, paths):
if pathid not in paths:
return Path(None, None, {})
if start not in paths[pathid].boxes or stop not in paths[pathid].boxes:
return Path(path.label, path.id, {})
if tracker in bidirectionaltrackers:
tracker = bidirectionaltrackers[tracker]()
return tracker.track(pathid, start, stop, basepath, paths)
return None
def online(tracker, start, stop, basepath, pathid, paths):
if pathid not in paths:
return Path(None, None, {})
if start not in paths[pathid].boxes:
return Path(path.label, path.id, {})
if tracker in onlinetrackers:
tracker = onlinetrackers[tracker]()
return tracker.track(pathid, start, stop, basepath, paths)
return None
def track(self, pathid, start, stop, basepath, paths):
bgs = cv2.BackgroundSubtractorMOG()
frames = getframes(basepath, False)
for frame in range(start, stop):
fgmask = bgs.apply(frames[frame])
cv2.imshow("Frame", fgmask)
cv2.waitKey(40)
cv2.destroyAllWindows()
return Path(path.label, path.id, {})
def track(self, pathid, start, stop, basepath, paths):
if pathid not in paths:
return Path(None, None, {})
path = paths[pathid]
if start not in path.boxes:
return Path(path.label, path.id, {})
startbox = path.boxes[start]
initialrect = (startbox.xtl, startbox.ytl, startbox.xbr-startbox.xtl, startbox.ybr-startbox.ytl)
frames = getframes(basepath, False)
previmage = frames[start]
imagesize = previmage.shape
prevpoints = np.array([])
points = getpoints(start, stop, frames, startbox)
boxes = meanshift(start, stop, points, initialrect, imagesize)
"""
for i in range(start, stop):
image = frames[i]
if i in points:
#cv2.circle(image, tuple(forwardmean[i,:]), 6, 0, 3)
for row in points[i]:
cv2.circle(image, tuple(row), 4, 0, 1)
if i in boxes:
box = boxes[i]
cv2.rectangle(image, (box.xtl,box.ytl), (box.xbr,box.ybr), 255,2)
cv2.imshow('Optical flow tracking', image)
cv2.waitKey(40)
"""
cv2.destroyAllWindows()
return Path(path.label, path.id, boxes)
def track(self, pathid, start, stop, basepath, paths):
path = paths[pathid]
if start not in path.boxes or stop not in path.boxes:
return Path(path.label, path.id, {})
startbox = path.boxes[start]
stopbox = path.boxes[stop]
initialrect = (startbox.xtl, startbox.ytl, startbox.xbr-startbox.xtl, startbox.ybr-startbox.ytl)
finalrect = (stopbox.xtl, stopbox.ytl, stopbox.xbr-stopbox.xtl, stopbox.ybr-stopbox.ytl)
frames = getframes(basepath, False)
previmage = frames[start]
imagesize = previmage.shape
forwardpoints = getpoints(start, stop, frames, startbox)
backwardpoints = getpoints(stop, start, frames, stopbox)
rowtoframe = sorted(list(set(forwardpoints.keys()) & set(backwardpoints.keys())))
forwardmean = np.array([np.mean(forwardpoints[frame], axis=0) for frame in rowtoframe])
backwardmean = np.array([np.mean(backwardpoints[frame], axis=0) for frame in rowtoframe])
meandiff = np.sum(np.square(forwardmean - backwardmean), axis=1)
mergeframe = rowtoframe[np.argmin(meandiff)]
print "Start frame", start
print "end frame", stop
print "Merge frame", mergeframe
startboxes = meanshift(start, mergeframe - 5, forwardpoints, initialrect, imagesize)
stopboxes = meanshift(stop, mergeframe + 5, backwardpoints, finalrect, imagesize)
boxes = {}
for row, frame in enumerate(rowtoframe):
if frame <= mergeframe and frame in startboxes:
boxes[frame] = startboxes[frame]
elif frame > mergeframe and frame in stopboxes:
boxes[frame] = stopboxes[frame]
"""
frametorow = {frame: row for row, frame in enumerate(rowtoframe)}
for i in range(start, stop):
image = frames[i]
#if i in forwardpoints:
#for row in forwardpoints[i]:
# cv2.circle(image, tuple(row), 4, 0, 1)
#if i in backwardpoints:
#cv2.circle(image, tuple(backwardmean[frametorow[i],:]), 6, 255, 3)
#for row in backwardpoints[i]:
# cv2.circle(image, tuple(row), 4, 255, 1)
if i in frametorow:
cv2.putText(image, str(meandiff[frametorow[i]]), (0, 30), cv2.FONT_HERSHEY_SIMPLEX, 1, 0, 2)
cv2.putText(image, "Frame" + str(i), (0, 60), cv2.FONT_HERSHEY_SIMPLEX, 1, 0, 2)
cv2.circle(image, tuple(forwardmean[frametorow[i],:]), 6, 0, 3)
cv2.circle(image, tuple(backwardmean[frametorow[i],:]), 6, 255, 3)
# Draw it on image
if i in boxes:
box = boxes[i]
if i < mergeframe:
cv2.rectangle(image, (box.xtl,box.ytl), (box.xbr,box.ybr), 0,2)
else:
cv2.rectangle(image, (box.xtl,box.ytl), (box.xbr,box.ybr), 255,2)
cv2.imshow('Optical flow tracking', image)
cv2.waitKey()
"""
cv2.destroyAllWindows()
return Path(path.label, path.id, boxes)