def process_response(q):
'''
This function defines the processes of the "Response Thread" that is
fully independent from the main thread. This one handles extracting
regions from the original (higher-res) frame and uploading the cropped
out portion locally or to the cloud.
'''
import boto3
s3 = boto3.resource("s3")
last_up = clock()
# initialize our largest detected area to 0
biggest_crop_area = 0
signal.signal(signal.SIGINT, signal.SIG_IGN)
while cap.isOpened():
# receive the data
data = q.get()
f = data["f"]
rects_sal = data["rects_sal"]
sz_scaled = data["sz_scaled"]
ts = data["ts"]
sz = getsize(f)
rr = (float(sz[0]) / sz_scaled[0], float(sz[1]) / sz_scaled[1])
# rescale the rectangular dimensions to our original resolution
bx = []
for x, y, w, h in rects_sal:
tmp = (x * rr[0], y * rr[1], w * rr[0], h * rr[1])
bx.append(tuple(int(x) for x in tmp))
if config.storage.save_cloud_crops or config.storage.save_local_crops:
if len(bx) > 0:
xx = tuple(
(min([min(x[0], x[0] + x[2]) for x in bx]), max([max(x[0], x[0] + x[2]) for x in bx])))
yy = tuple(
(min([min(x[1], x[1] + x[3]) for x in bx]), max([max(x[1], x[1] + x[3]) for x in bx])))
# only continue if the area of is not zero
if abs(yy[0] - yy[1]) > 0 and abs(xx[0] - xx[1]) > 0:
f_mask = f[min(yy):max(yy), min(xx):max(xx)]
cropped_area = (max(xx) - min(xx)) * (max(yy) - min(yy))
if (clock() - last_up) > config.storage.min_upload_delay:
biggest_crop_area = 0
# Always send the frames that contain detected people/faces.
# If detecting people/faces is disabled in the config, it
# is not affected.
if (cropped_area > biggest_crop_area) or data["num_bodies"] > 0 or data["num_faces"] > 0:
biggest_crop_area = cropped_area
root_path = join(
join(sDir, "cropped-regions"), date_pretty())
fn = root_path + "_regions.jpg"
res, img = cv2.imencode(
".jpg", f_mask, [int(cv2.IMWRITE_JPEG_QUALITY), config.storage.encoding_quality])
if res and config.storage.save_local_crops:
print("saving frame locally: {}".format(root_path))
# todo: save locally
if res and config.storage.save_cloud_crops:
last_up = clock()
img = img.tostring()
print(
"uploading frame to s3: {}".format(basename(fn)))
print(
"-- time since last upload: {}s".format(clock() - last_up))
s3.Object("cloudguard-in",
basename(fn)).put(Body=img,
Metadata={"Content-Type": "Image/jpeg",
"Number-Detected-Motion": str(data["num_motion"]),
"Number-Detected-Bodies": str(data["num_bodies"]),
"Number-Detected-Faces": str(data["num_faces"]),
"Captured-Timestamp": str(ts),
"Captured-Timestamp-Timezone": "UTC"})
last_frame_time = t
ts = datetime.utcnow()
task = pool.apply_async(process_motion_frame, args=(
q, f, t, ts, MFA, config.camera.rot, config.computing.width))
pending.append(task)
# save a local video stream of detection motion if enabled
if config.storage.save_local_vids:
# motion has just begun, generate a new filename and write the
# first frame
if (datetime.utcnow() - LMT) < timedelta(seconds=config.computing.last_motion_timeout):
if MFA is False:
MFA = True
streamId += 1
vwParams["filename"] = join(
vsDir, str(vWfn[0] + "_{:04d}_".format(streamId) + date_pretty() + vWfn[1]))
vW = cv2.VideoWriter(**vwParams)
# no activity
else:
if MFA is True:
MFA = False
vW = None
if config.window.enabled:
# update the slider values to the background model and
# bind tasks to keystrokes if the window is enabled
# only with opencv3
if imutils.is_cv3():
# refresh the background subtraction parameters
bgSh = cv2.getTrackbarPos('Motion Hist.', config.window.name)
