def main():
# Register exit handler
signal.signal(signal.SIGINT, signal_handler)
# print('Press Ctrl+C to exit.')
# Load the cameras configuration
data = yaml.load(open(Config.CAMS_YML, 'r'))
cams = data['cams'] # type: dict
print("Loaded {}".format(Config.CAMS_YML))
# Connect to the redis instance
rdb = redis.StrictRedis(host=Config.REDIS_HOST, port=Config.REDIS_PORT, db=Config.REDIS_DB, decode_responses=True)
# TODO: Consider whether we should read in some other way.
# Clear keys so that the stats are right.
for key in rdb.scan_iter("{}:*".format(Config.REDIS_PREFIX)):
# print("Deleting: {}".format(key))
rdb.delete(key)
# Create every cam feeder
for cam_name, cam in cams.items():
print('Adding cam {0} to the dict'.format(cam_name))
cam_url = cam.get('cam_url')
fps = cam.get('fps')
queue = Queue()
p = multiprocessing.Process(target=producer, args=(rdb, queue, cam_name, cam_url, fps))
p.start()
if os.path.isdir(cam_url):
while True:
lst_file = os.listdir(cam_url)
for file in lst_file:
out_of_memory()
file_url = os.path.join(cam_url, file)
queue.put((-1, file))
cf = image2pipe.images_from_url(q=queue, video_url=file_url, fps=fps, scale=(1920, 1080))
cf.start()
while cf.is_alive() is True:
gevent.sleep(1)
os.remove(file_url)
print('Wait for new file')
gevent.sleep(5)
elif os.path.isfile(cam_url):
queue.put((-1, cam_url))
cf = image2pipe.images_from_url(q=queue, video_url=cam_url, fps=fps, scale=(1920, 1080))
cf.start()
watchdog(rdb)
def test():
Path("tmp").mkdir(exist_ok=True)
q = Queue(maxsize=4)
# equivalent to
# ffmpeg -v error -ss 00:00:00 -i data/COVID-19.mp4 -an -sn -f image2pipe -vcodec rawvideo -pix_fmt rgb24 -vf fps=2,scale=640x360
decoder = image2pipe.images_from_url(q,
"data/COVID-19.mp4",
fps="2",
scale=(640, 360),
pix_fmt='rgb24')
# if this is the press conference we might be scaling a (1280, 720) -> to (640, 360) here
decoder.start()
face_confidence = 0
for pair in yield_from_queue(q):
i, img = pair
timer_start = timer()
image = Image.fromarray(img)
is_face, confidence = face_in_image(image)
timer_total = timer() - timer_start
print("Checking for faces took {0:.3f}s.".format(timer_total))
if is_face:
image.save("tmp/face_test_{}.jpg".format(i))
face_confidence += 1
if face_confidence > 10:
break
else:
face_confidence = 0
print("{} {}:{}".format(i, is_face, confidence))
def test_vf(self):
q = Queue()
# decoder = image2pipe.images_from_url(q, VIDEO_URL, fps="30", scale=SCALE, vf=["cropdetect=24:16:0"])
decoder = image2pipe.images_from_url(q, VIDEO_URL, fps="30", scale=SCALE, vf=["crop=224:224:0:36"])
decoder.start()
fn, img = q.get()
cv2.imshow("frame %d" % 0, img)
cv2.waitKey()
cv2.destroyAllWindows()
def test_rgb24_from_url(self):
q = Queue()
decoder = image2pipe.images_from_url(q, VIDEO_URL, fps="30", scale=SCALE)
decoder.start()
for i in range(30):
fn, img = q.get()
cv2.imshow("frame %d" % i, img)
cv2.waitKey()
cv2.destroyAllWindows()
def test_min_params(self):
q = Queue()
decoder = image2pipe.images_from_url(q, VIDEO_URL)
decoder.start()
for i in range(30):
fn, img = q.get()
cv2.imshow("frame %d" % i, img)
cv2.waitKey()
cv2.destroyAllWindows()
def test_stitch(self):
fps = "30"
out_url = "out.ts"
scale = (1000, 552)
bgr_q = Queue()
decoder = image2pipe.images_from_url(bgr_q, VIDEO_URL, fps="30", scale=(1000, 552))
decoder.start()
FORMAT_MPEGTS = "mpegts"
rtmpt = image2pipe.StitchVideoProcess(bgr_q, out_url, fps, scale, FORMAT_MPEGTS)
rtmpt.start()
rtmpt.join()
def __init__(self,
filename: str,
frame_range: Optional[Tuple[int, int]] = None,
process_nth_frames: Optional[int] = None,
min_size: Optional[int] = None,
video_info: Optional[VideoInfo] = None):
if video_info is not None:
frame_count, width, height, fps = video_info
self.frame_count = frame_count
self.width = width
self.height = height
self.fps = fps
else:
info = ffprobe(filename)['streams'][0]
frame_count = int(info['nb_frames'])
self.frame_count = frame_count
width = int(info['width'])
height = int(info['height'])
self.width = width
self.height = height
fps = Fraction(info['avg_frame_rate'])
self.fps = float(fps)
self.queue = Queue()
if frame_range is not None:
frame_index_start, frame_index_end = frame_range
if frame_index_end == -1:
frame_index_end = frame_count
if frame_index_start < 0:
print(
f"WARNING! Invalid frame_range {frame_range}. frame_index_start cant be less then 0. Using default value of 0 instead."
)
frame_index_start = 0
if frame_index_end > frame_count:
print(
f"WARNING! Invalid frame_range {frame_range}. frame_index_end cant be greater then {frame_count }. Using default value of {frame_count - 1} instead."
)
frame_index_end = frame_count
else:
frame_index_start = 0
frame_index_end = frame_count
ss = str(float(frame_index_start / fps))
# to = str(float(frame_index_end / fps))
self._frame_index_end = frame_index_end
self._frame_index_start = frame_index_start
if process_nth_frames:
fps /= process_nth_frames
if min_size is not None:
if width < height:
scale = (min_size, int(round(height / (width / min_size))))
else:
scale = (int(round(width / (height / min_size))), min_size)
else:
scale = (width, height)
decoder = image2pipe.images_from_url(self.queue,
filename,
ss=ss,
fps=fps,
scale=scale,
buffer_size=scale)
decoder.start()
if (invert):
temp = cv2.bitwise_not(temp)
return temp
def crop(image, coordinates):
x_start = coordinates[2]
x_end = x_start + coordinates[0]
y_start = coordinates[3]
y_end = y_start + coordinates[1]
return image[y_start:y_end, x_start:x_end]
q = Queue()
decoder = image2pipe.images_from_url(q,
sys.argv[1],
fps=sys.argv[2],
scale=(1920, 1080))
decoder.start()
with open(sys.argv[3], "a") as out:
sys.stdout = out
for frame_number, image in image2pipe.utils.yield_from_queue(q):
tesseract = tesserpy.Tesseract("/usr/share/tesseract-ocr/tessdata",
language="eng",
oem=tesserpy.OEM_TESSERACT_ONLY)
tesseract.tessedit_pageseg_mode = tesserpy.PSM_SINGLE_LINE
frametime = datetime.utcnow().isoformat()
print(frametime + " #" + str(frame_number), end=' ')
for datum, settings in crops.items():
dir(settings)
cropped = crop(image, settings['coords'])
# if args.read_path is None or args.save_path is None:
# parser.print_usage()
# sys.exit(1)
# use CUDA?
args.cuda = not args.no_gpu and torch.cuda.is_available()
frames_q = None
if args.read_path.find("://") > -1:
print("Decoding input video %s" % args.read_path)
frames_q = Queue(maxsize=NUM_IN_FR_EXT * 5)
if args.read_path.find(":/") > -1:
probe = image2pipe.ffprobe(args.read_path)
decoder = image2pipe.images_from_url(frames_q,
args.read_path,
scale=None)
decoder.start()
# If save_path does not exist, create it
if not os.path.exists(args.save_path):
os.makedirs(args.save_path)
# Sets data type according to CPU or GPU modes
if args.cuda:
device = torch.device('cuda')
else:
device = torch.device('cpu')
# Create models
print('Loading model ...')