mask = cv2.imread(mask_path, 0)
vp_path = '../data/gt/2016-ITS-BrnoCompSpeed/results/{}/system_dubska_bmvc14.json'.format(
session_id)
with open(vp_path, 'r') as f_buff:
"""
vp has 2 keys
- cars, list of cars detected, its frame and posX and posY
- camera_calibration, the calibration parameter result (pp, vp1, and vp2)
"""
vp = json.load(f_buff)
vp1 = vp['camera_calibration']['vp1']
vp2 = vp['camera_calibration']['vp2']
fi = FrameIterator('../data/sample/{}'.format(session_id))
bg = BackgroundModel(fi)
print("Learning")
bg.learn(tot_frame_init=1)
print("Done")
ctr_frame = 0
# what I want is the function to find most left tangen and right tangen of
# given blob list and a point
while True:
img = next(fi)
frame = img
fg = bg.apply(img)
# remove shadows, i.e value 127
"""
vp = json.load (f_buff)
vps[_id][view] = {
'vp1' : vp['camera_calibration']['vp1'],
'vp2' : vp['camera_calibration']['vp2']
}
# generate frame iterator
fi = {_id : {}}
for view in session[_id] :
fi[_id][view] = FrameIterator ('../data/sync_25fps/session{}_{}'.format (_id, view))
# define background model
bms = {_id : {}}
for view in session[_id] :
bms[_id][view] = BackgroundModel (fi[_id][view])
print ("Learning for session {}-{}".format (_id, view))
bms[_id][view].learn (tot_frame_init=2)
print ("Done")
# initialing prev blobs
prev_fg = {_id : {}}
for view in session[_id] :
prev_fg[_id][view] = [None, None]
for i in range (2) :
img = next (fi[_id][view])
# by background subtraction
fg = bms[_id][view].apply (img)
# remove shadows, i.e value 127
fg = cv2.threshold (fg, 200, 255, cv2.THRESH_BINARY)[1]
masks = {}
HEIGHT = 400 # constant height
for view in VIEW :
img = cv2.imread (img_path.format (ses_id, view), 1)
points = GT['session{}'.format (ses_id)][view]
corner = get_corner_ground (vp[view]['vp1'], vp[view]['vp2'], points)
# get rectangular homography mapping
corner_gt = np.float32 (corner)
corner_wrap = np.float32 ([[0,300],[0,0], [1000,0], [1000, 300]])
M[view] = cv2.getPerspectiveTransform (corner_gt, corner_wrap)
# background subtraction
bms[view] = BackgroundModel (fi[view], detectShadows=False)
bms[view].learn (tot_frame_init=2)
# for 3 frame difference
prev_img[view] = [None, None]
for i in range (2) :
img = next (fi[view])
img_color = img.copy ()
img = cv2.cvtColor (img, cv2.COLOR_BGR2GRAY)
# save background
prev_img[view][i] = None
mask_path = '../../data/gt/2016-ITS-BrnoCompSpeed/dataset/session{}_{}/video_mask.png'.format (ses_id, view)
masks[view] = cv2.imread (mask_path, 0)
img = next(fi[view])
img = cv2.cvtColor(img, cv2.COLOR_BGR2GRAY)
dst = cv2.warpPerspective(img, M[view], (1000, 300))
# save background
prev_img[i] = img
# save tsi
prev_tsi[i] = tsi_object[view].apply(img)
prev_epi[i] = dst[70:80:, :]
# fdiff_view[view] = FrameDifference (*prev_img)
fdiff_tsi[view] = FrameDifference(*prev_tsi)
fdiff_epi[view] = FrameDifference(*prev_epi)
bm_epi[view] = BackgroundModel(iter(prev_epi))
bm_epi[view].learn(tot_frame_init=2)
mask_path = '../../data/gt/2016-ITS-BrnoCompSpeed/dataset/session{}_{}/video_mask.png'.format(
ses_id, view)
masks[view] = cv2.imread(mask_path, 0)
"""
-------------------------------------------------------------------------------
Main Program
-------------------------------------------------------------------------------
"""
# what I want is a function to generate 3D using Vanishing Point (it already exist)
ctr = 0
while True:
ctr += 1
