def cleanOlds(self, frame):
track_temp = []
#for trk in self.tracklets:
for trk in self.tracklets:
life = frame - sorted(trk.points.keys())[-1]
if life > self.vanish_time:
# saving in file
if len(trk.points) >= self.minlen:
name = self.name + '_' + str(trk.id) + self.token
line = ''
for frame_point in sorted(trk.points):
line += '%06d,' % frame_point
line += str(trk.points[frame_point][0]) + ' '
line += str(trk.points[frame_point][1]) + '\n'
u_save2File(name, line)
self.files.append(name)
else:
track_temp.append(trk)
# setting to False each tracklet for Kalman automatick tracking
trk.visited = False
self.tracklets = track_temp
def detect_image_file(info, params, model_params):
# generate image with body parts
file_name = info['file']
file_out = info['out']
oriImg = cv2.imread(file_name) # B,G,R order
canvas, ans = process(oriImg, params, model_params)
cv2.imwrite(file_out, canvas)
u_save2File(os.path.splitext(file_out)[0] + '.txt', ans)
def dump(self):
for trk in self.tracklets:
if len(trk.points) >= self.minlen:
name = self.name + '_' + str(trk.id) + self.token
line = ''
for frame_point in sorted(trk.points):
line += '%06d,' % frame_point
line += str(trk.points[frame_point][0]) + ' '
line += str(trk.points[frame_point][1]) + '\n'
u_save2File(name, line)
self.files.append(name)
def detect_video_file(info, params, model_params):
file_name = info['file']
ini = info['ini']
fin = info['fin']
out_folder = info['out_folder']
visual = info['visual']
prop = info['prop']
supported = info['supported']
step = info['step']
data = {}
out_prop_folder = out_folder + '/props'
base = os.path.basename(file_name)
base = os.path.splitext(base)[0] + '.prop'
name = out_prop_folder + '/' + base
if os.path.isfile(name):
print(' File previously processed ')
return
if not os.path.exists(out_folder):
os.makedirs(out_folder)
if step == 1:
video = video_sequence_by1(file_name, ini, fin)
else:
video = video_sequence_byn(file_name, step, ini, fin)
cap = cv2.VideoCapture(file_name)
ret, frame = cap.read()
print('Reading :' + file_name)
# Iamge video generation ...................................................
if visual > 0:
if visual == 1:
nframe = 0
while (ret):
name = out_folder + '/%05d' % (nframe) + '.png'
print('Save: ', name)
params_c = copy.deepcopy(params)
model_params_c = copy.deepcopy(model_params)
canvas, ans = process(frame, params_c, model_params_c)
cv2.imwrite(name, canvas)
#ret, frame = video.getCurrent()
ret, frame = cap.read()
nframe += 1
# Video Pose image generation...........................................
# visual = 2
else:
fourcc = cv2.VideoWriter_fourcc(*'XVID')
file_n = os.path.basename(file_name)
file_n = os.path.splitext(file_n)[0] + '.avi'
name_v = out_folder + '/' + file_n
fps = int(video.cap.get(cv2.CAP_PROP_FPS))
out = cv2.VideoWriter(name_v, fourcc, fps,
(int(video.width), int(video.height)))
pos = 1
while (ret):
print(pos)
params_c = copy.deepcopy(params)
model_params_c = copy.deepcopy(model_params)
canvas, ans = process(frame, params_c, model_params_c)
out.write(canvas)
#print('frame')
#print(video.current-1)
#ret, frame = video.getCurrent()
ret, frame = cap.read()
pos += 1
print('Save video in: ', name_v)
data['out_video_file'] = name_v
out.release()
else:
# Visual = 0
# Tracklet file generation.....................................................
print('Tracklet generation only')
final_ans = ''
# if you have previous human detection
if supported:
frame_list = []
detection_file = os.path.splitext(file_name)[0] + '.txt'
file = open(detection_file, 'r')
for line in file:
if len(line) < 1:
continue
split_line = line.split(' ')
frame_list.append(int(split_line[0]))
frame_list = sorted(list(set(frame_list)))
#print(frame_list)
ret, frame = video.getCurrent()
while (ret):
#print (video.current)
if (video.current - 1) in frame_list:
#name = out_folder + '/%05d' % (video.current - 1) + '.png'
#print('Save: ', name)
params_c = copy.deepcopy(params)
model_params_c = copy.deepcopy(model_params)
canvas, ans = process(frame, params_c, model_params_c)
if ans is not '\n':
final_ans = final_ans + '%05d-' % (video.current -
1) + ans
#cv2.imwrite(name, canvas)
#print(final_ans)
#cv2.imshow('frame', canvas)
#if cv2.waitKey(1) & 0xFF == ord('q'):
# break
ret, frame = video.getCurrent()
#print (final_ans)
base = os.path.basename(file_name)
base = os.path.splitext(base)[0] + '.txt'
name = out_folder + '/' + base
u_save2File(name, final_ans)
data['tracklet_file'] = name
# in complete analisys
else:
while (ret):
params_c = copy.deepcopy(params)
model_params_c = copy.deepcopy(model_params)
canvas, ans = process(frame, params_c, model_params_c)
if ans is not '\n':
final_ans = final_ans + '%05d-' % (video.current - 1) + ans
ret, frame = video.getCurrent()
base = os.path.basename(file_name)
base = os.path.splitext(base)[0] + '.txt'
name = out_folder + '/' + base
u_save2File(name, final_ans)
data['tracklet_file'] = name
#............................................................................
#proerty flag
if prop:
out_prop_folder = out_folder + '/props'
if not os.path.exists(out_prop_folder):
os.makedirs(out_prop_folder)
data_ = {
"video_file": file_name,
"ini": video.pos_ini,
"fin": video.pos_fin,
"width": video.width,
"height": video.height
}
data.update(data_)
base = os.path.basename(file_name)
base = os.path.splitext(base)[0] + '.prop'
name = out_prop_folder + '/' + base
#saving in file
print('Save prop in: ', name)
with open(name, 'w') as outfile:
json.dump(data, outfile)