def process_live_feed(self, cv_image, objects, distancings, violating_objects, out, out_birdseye):
dist_threshold = float(self.config.get_section_dict("PostProcessor")["DistThreshold"])
birds_eye_window = np.zeros(self.birds_eye_resolution[::-1] + (3,), dtype="uint8")
class_id = int(self.config.get_section_dict('Detector')['ClassID'])
output_dict = visualization_utils.visualization_preparation(objects, distancings, dist_threshold)
category_index = {class_id: {
"id": class_id,
"name": "Pedestrian",
}} # TODO: json file for detector config
face_index = {
0: "YES",
1: "NO",
-1: "N/A",
}
# Draw bounding boxes and other visualization factors on input_frame
visualization_utils.visualize_boxes_and_labels_on_image_array(
cv_image,
output_dict["detection_boxes"],
output_dict["detection_classes"],
output_dict["detection_scores"],
output_dict["detection_colors"],
category_index,
instance_masks=output_dict.get("detection_masks"),
use_normalized_coordinates=True,
line_thickness=3,
face_labels=output_dict["face_labels"],
face_index=face_index
)
# TODO: Implement perspective view for objects
birds_eye_window = visualization_utils.birds_eye_view(
birds_eye_window, output_dict["detection_boxes"], output_dict["violating_objects"])
try:
fps = self.detector.fps
except:
# fps is not implemented for the detector instance"
fps = None
# Put fps to the frame
# region
# -_- -_- -_- -_- -_- -_- -_- -_- -_- -_- -_- -_- -_- -_-
txt_fps = 'Frames rate = ' + str(fps) + '(fps)' # Frames rate = 95 (fps)
# (0, 0) is the top-left (x,y); normalized number between 0-1
origin = (0.05, 0.93)
visualization_utils.text_putter(cv_image, txt_fps, origin)
# -_- -_- -_- -_- -_- -_- -_- -_- -_- -_- -_- -_- -_- -_-
# endregion
# Put environment score to the frame
# region
# -_- -_- -_- -_- -_- -_- -_- -_- -_- -_- -_- -_- -_- -_-
env_score = mx_environment_scoring_consider_crowd(len(objects), len(violating_objects))
txt_env_score = 'Env Score = ' + str(env_score) # Env Score = 0.7
origin = (0.05, 0.98)
visualization_utils.text_putter(cv_image, txt_env_score, origin)
# -_- -_- -_- -_- -_- -_- -_- -_- -_- -_- -_- -_- -_- -_-
# endregion
out.write(cv_image)
out_birdseye.write(birds_eye_window)
def update(self, input_frame, nn_out):
"""
Args:
input_frame: uint8 numpy array with shape (img_height, img_width, 3)
nn_out: List of dicionary contains normalized numbers of bounding boxes
{'id' : '0-0', 'bbox' : [x0, y0, x1, y1], 'score' : 0.99(optional} of shape [N, 3] or [N, 2]
Returns:
draw the bounding boxes to an output frame
"""
# Get a proper dictionary of bounding boxes and colors for visualizing_boxes_and_labels_on_image_array function
output_dict = vis_util.visualization_preparation(nn_out) # TODO
#print(output_dict)
category_index = {
0: {
"id": 0,
"name": "Mask",
},
1: {
"id": 1,
"name": "Face",
}}
# Draw bounding boxes and other visualization factors on input_frame
vis_util.visualize_boxes_and_labels_on_image_array(
input_frame,
output_dict["detection_boxes"],
output_dict["detection_classes"],
output_dict["detection_scores"],
output_dict["detection_colors"],
category_index,
instance_masks=output_dict.get("detection_masks"),
use_normalized_coordinates=True,
line_thickness=3,
)
# TODO: Implement perspective view for objects
try:
self._displayed_items['fps'] = self.__ENGINE_INSTANCE.detector.fps
except:
# fps is not implemented for the detector instance"
self._displayed_items['fps'] = None
# Put fps to the frame
# region
# -_- -_- -_- -_- -_- -_- -_- -_- -_- -_- -_- -_- -_- -_-
txt_fps = 'Frames rate = ' + str(self._displayed_items['fps']) + '(fps)' # Frames rate = 95 (fps)
# (0, 0) is the top-left (x,y); normalized number between 0-1
origin = (0.05, 0.93)
vis_util.text_putter(input_frame, txt_fps, origin)
# -_- -_- -_- -_- -_- -_- -_- -_- -_- -_- -_- -_- -_- -_-
# endregion
# Lock the main thread and copy input_frame to output_frame
with self._lock:
self._output_frame = input_frame.copy()
def update(self, cv_image, objects, post_processing_data, fps, log_time):
if not self.live_feed_enabled:
return
self.update_history(post_processing_data["tracks"])
distancings = post_processing_data.get("distances", [])
dist_threshold = post_processing_data.get("dist_threshold", 0)
birds_eye_window = np.zeros(self.birds_eye_resolution[::-1] + (3, ),
dtype="uint8")
class_id = int(
get_model_json_file_or_return_default_values(
self.config,
self.config.get_section_dict('Detector')['Device'],
self.camera_id)["variables"]["ClassID"])
roi_contour = ObjectsFilteringPostProcessor.get_roi_contour(
self.roi_file_path)
if roi_contour is not None:
color = (41, 127, 255) # #ff7f29 (255, 127, 41)
visualization_utils.draw_contour(cv_image, roi_contour, color)
output_dict = visualization_utils.visualization_preparation(
objects, distancings, dist_threshold)
category_index = {
class_id: {
"id": class_id,
"name": "Pedestrian",
}
} # TODO: json file for detector config
face_index = {
0: "YES",
1: "NO",
-1: "N/A",
}
# Assign object's color to corresponding track history
for i, track_id in enumerate(output_dict["track_ids"]):
self.track_hist[track_id][1].append(
output_dict["detection_colors"][i])
# Draw bounding boxes and other visualization factors on input_frame
visualization_utils.visualize_boxes_and_labels_on_image_array(
cv_image,
output_dict["detection_boxes"],
output_dict["detection_classes"],
output_dict["detection_scores"],
output_dict["detection_colors"],
output_dict["track_ids"],
category_index,
instance_masks=output_dict.get("detection_masks"),
use_normalized_coordinates=True,
line_thickness=3,
face_labels=output_dict["face_labels"],
face_index=face_index)
# TODO: Implement perspective view for objects
birds_eye_window = visualization_utils.birds_eye_view(
birds_eye_window, output_dict["detection_boxes"],
output_dict["violating_objects"])
# Put occupancy to the frame
# region
# -_- -_- -_- -_- -_- -_- -_- -_- -_- -_- -_- -_- -_- -_-
txt_fps = 'Occupancy = ' + str(len(objects))
# (0, 0) is the top-left (x,y); normalized number between 0-1
origin = (0.05, 0.93)
visualization_utils.text_putter(cv_image, txt_fps, origin)
# -_- -_- -_- -_- -_- -_- -_- -_- -_- -_- -_- -_- -_- -_-
# endregion
# visualize tracks
# region
# -_- -_- -_- -_- -_- -_- -_- -_- -_- -_- -_- -_- -_- -_-
visualization_utils.draw_tracks(cv_image,
self.track_hist,
radius=1,
thickness=1)
# -_- -_- -_- -_- -_- -_- -_- -_- -_- -_- -_- -_- -_- -_-
# endregion
self.out.write(cv_image)
self.out_birdseye.write(birds_eye_window)
def update(self, cv_image, objects, post_processing_data, fps):
if not self.live_feed_enabled:
return
self.update_history(post_processing_data["tracks"])
distancings = post_processing_data.get("distances", [])
violating_objects = post_processing_data.get("violating_objects", [])
dist_threshold = post_processing_data.get("dist_threshold", 0)
birds_eye_window = np.zeros(self.birds_eye_resolution[::-1] + (3,), dtype="uint8")
class_id = int(self.config.get_section_dict('Detector')['ClassID'])
output_dict = visualization_utils.visualization_preparation(objects, distancings, dist_threshold)
category_index = {class_id: {
"id": class_id,
"name": "Pedestrian",
}} # TODO: json file for detector config
face_index = {
0: "YES",
1: "NO",
-1: "N/A",
}
# Assign object's color to corresponding track history
for i, track_id in enumerate(output_dict["track_ids"]):
self.track_hist[track_id][1].append(output_dict["detection_colors"][i])
# Draw bounding boxes and other visualization factors on input_frame
visualization_utils.visualize_boxes_and_labels_on_image_array(
cv_image,
output_dict["detection_boxes"],
output_dict["detection_classes"],
output_dict["detection_scores"],
output_dict["detection_colors"],
output_dict["track_ids"],
category_index,
instance_masks=output_dict.get("detection_masks"),
use_normalized_coordinates=True,
line_thickness=3,
face_labels=output_dict["face_labels"],
face_index=face_index
)
# TODO: Implement perspective view for objects
birds_eye_window = visualization_utils.birds_eye_view(
birds_eye_window, output_dict["detection_boxes"], output_dict["violating_objects"])
# Put fps to the frame
# region
# -_- -_- -_- -_- -_- -_- -_- -_- -_- -_- -_- -_- -_- -_-
txt_fps = 'Frames rate = ' + str(fps) + '(fps)' # Frames rate = 95 (fps)
# (0, 0) is the top-left (x,y); normalized number between 0-1
origin = (0.05, 0.93)
visualization_utils.text_putter(cv_image, txt_fps, origin)
# -_- -_- -_- -_- -_- -_- -_- -_- -_- -_- -_- -_- -_- -_-
# endregion
# Put environment score to the frame
# region
# -_- -_- -_- -_- -_- -_- -_- -_- -_- -_- -_- -_- -_- -_-
env_score = mx_environment_scoring_consider_crowd(len(objects), len(violating_objects))
txt_env_score = 'Env Score = ' + str(env_score) # Env Score = 0.7
origin = (0.05, 0.98)
visualization_utils.text_putter(cv_image, txt_env_score, origin)
# -_- -_- -_- -_- -_- -_- -_- -_- -_- -_- -_- -_- -_- -_-
# endregion
# visualize tracks
# region
# -_- -_- -_- -_- -_- -_- -_- -_- -_- -_- -_- -_- -_- -_-
visualization_utils.draw_tracks(cv_image, self.track_hist, radius=1, thickness=1)
# -_- -_- -_- -_- -_- -_- -_- -_- -_- -_- -_- -_- -_- -_-
# endregion
self.out.write(cv_image)
self.out_birdseye.write(birds_eye_window)
def process_video(self, video_uri):
if self.device == 'Jetson':
from libs.detectors.jetson.detector import Detector
self.detector = Detector(self.config)
elif self.device == 'EdgeTPU':
from libs.detectors.edgetpu.detector import Detector
self.detector = Detector(self.config)
elif self.device == 'Dummy':
from libs.detectors.dummy.detector import Detector
self.detector = Detector(self.config)
elif self.device == 'x86':
from libs.detectors.x86.detector import Detector
self.detector = Detector(self.config)
if self.device != 'Dummy':
print('Device is: ', self.device)
print('Detector is: ', self.detector.name)
print('image size: ', self.image_size)
input_cap = cv.VideoCapture(video_uri)
fps = input_cap.get(cv.CAP_PROP_FPS)
if (input_cap.isOpened()):
logger.info(f'opened video {video_uri}')
else:
logger.error(f'failed to load video {video_uri}')
return
self.running_video = True
# enable logging gstreamer Errors (https://stackoverflow.com/questions/3298934/how-do-i-view-gstreamer-debug-output)
os.environ['GST_DEBUG'] = "*:1"
out, out_birdseye = (
self.gstreamer_writer(feed, fps, resolution)
for (feed, resolution) in (('default', self.resolution), (
'default-birdseye',
self.birds_eye_resolution)) # TODO: use camera-id
)
dist_threshold = float(
self.config.get_section_dict("PostProcessor")["DistThreshold"])
class_id = int(self.config.get_section_dict('Detector')['ClassID'])
frame_num = 0
while input_cap.isOpened() and self.running_video:
_, cv_image = input_cap.read()
birds_eye_window = np.zeros(self.birds_eye_resolution[::-1] +
(3, ),
dtype="uint8")
if np.shape(cv_image) != ():
cv_image, objects, distancings = self.__process(cv_image)
output_dict = visualization_utils.visualization_preparation(
objects, distancings, dist_threshold)
category_index = {
class_id: {
"id": class_id,
"name": "Pedestrian",
}
} # TODO: json file for detector config
# Draw bounding boxes and other visualization factors on input_frame
visualization_utils.visualize_boxes_and_labels_on_image_array(
cv_image,
output_dict["detection_boxes"],
output_dict["detection_classes"],
output_dict["detection_scores"],
output_dict["detection_colors"],
category_index,
instance_masks=output_dict.get("detection_masks"),
use_normalized_coordinates=True,
line_thickness=3,
)
# TODO: Implement perspective view for objects
birds_eye_window = visualization_utils.birds_eye_view(
birds_eye_window, output_dict["detection_boxes"],
output_dict["violating_objects"])
try:
fps = self.detector.fps
except:
# fps is not implemented for the detector instance"
fps = None
# Put fps to the frame
# region
# -_- -_- -_- -_- -_- -_- -_- -_- -_- -_- -_- -_- -_- -_-
txt_fps = 'Frames rate = ' + str(
fps) + '(fps)' # Frames rate = 95 (fps)
# (0, 0) is the top-left (x,y); normalized number between 0-1
origin = (0.05, 0.93)
visualization_utils.text_putter(cv_image, txt_fps, origin)
# -_- -_- -_- -_- -_- -_- -_- -_- -_- -_- -_- -_- -_- -_-
# endregion
# Put environment score to the frame
# region
# -_- -_- -_- -_- -_- -_- -_- -_- -_- -_- -_- -_- -_- -_-
violating_objects = extract_violating_objects(
distancings, dist_threshold)
env_score = mx_environment_scoring_consider_crowd(
len(objects), len(violating_objects))
txt_env_score = 'Env Score = ' + str(
env_score) # Env Score = 0.7
origin = (0.05, 0.98)
visualization_utils.text_putter(cv_image, txt_env_score,
origin)
# -_- -_- -_- -_- -_- -_- -_- -_- -_- -_- -_- -_- -_- -_-
# endregion
out.write(cv_image)
out_birdseye.write(birds_eye_window)
frame_num += 1
if frame_num % 10 == 1:
logger.info(f'processed frame {frame_num} for {video_uri}')
else:
continue
self.logger.update(objects, distancings)
input_cap.release()
out.release()
out_birdseye.release()
del self.detector
self.running_video = False