def detect_objects(threshold = 0.1, top_count=3):
interpreter = common.make_interpreter(default_model)
interpreter.allocate_tensors()
labels = load_labels(default_labels)
cap = cv2.VideoCapture(default_camera_idx)
if cap.isOpened():
for i in range(0,15):
ret, frame = cap.read()
time.sleep(1/1000)
if not ret:
break
cv2_im = frame
cv2_im_rgb = cv2.cvtColor(cv2_im, cv2.COLOR_BGR2RGB)
pil_im = Image.fromarray(cv2_im_rgb)
common.set_input(interpreter, pil_im)
interpreter.invoke()
objs = get_output(interpreter, score_threshold=threshold, top_k=top_count)
cv2_im = append_objs_to_img(cv2_im, objs, labels)
cv2.imshow('detect', cv2_im)
cv2.waitKey(50)
def make(obj):
return Result(
percent = int(100 * obj.score),
label = labels.get(obj.id, 'unknown')
)
cap.release()
return [make(obj) for obj in objs]
def __init__(self):
print('Loading {} with {} labels.'.format(model, labels))
self.interpreter = common.make_interpreter(model)
self.interpreter.allocate_tensors()
self.labels = load_labels(labels)
self.video = cv2.VideoCapture(0)
self.file = open("/home/mendel/person_detected.txt","w")
def main():
parser = argparse.ArgumentParser(
formatter_class=argparse.ArgumentDefaultsHelpFormatter)
parser.add_argument(
'--model', help='File path of .tflite model.', default='inception_v4_299_quant_edgetpu.tflite')
parser.add_argument(
'--labels', help='File path of labels file.', default='imagenet_labels.txt')
parser.add_argument(
'--top_k', help='Number of classifications to list', type=int, default=1)
args = parser.parse_args()
print('Initializing TF Lite interpreter...')
interpreter = common.make_interpreter(os.path.join(default_model_dir,args.model))
interpreter.allocate_tensors()
labels = load_labels(os.path.join(default_model_dir, args.labels))
cap = cv2.VideoCapture(0)
while (True):
ret, frame = cap.read()
cv2_im_rgb = cv2.cvtColor(frame, cv2.COLOR_BGR2RGB)
pil_im = Image.fromarray(cv2_im_rgb)
common.set_input(interpreter, pil_im)
results = classify_image(interpreter, pil_im, args.top_k)
for label_id, prob in results:
cv2.putText(frame, labels[label_id], (5,35), cv2.FONT_HERSHEY_SIMPLEX, .7, (0,0,0), 2)
print('%s: %.5f' % (labels[label_id], prob))
cv2.imshow('Classification', frame)
if cv2.waitKey(1) == ord('q'):
break
cap.release()
cv2.destroyAllWindows()
def detect_objects(args):
interpreter = common.make_interpreter(args.model)
interpreter.allocate_tensors()
labels = load_labels(args.labels)
dirname = args.images
dirpath = Path('results/' + dirname)
if dirpath.exists() and dirpath.is_dir():
shutil.rmtree(dirpath)
Path("results/" + dirname).mkdir(parents=True, exist_ok=True)
for filename in glob.glob(dirname + "/*.jpeg"):
print(filename)
name = os.path.basename(filename)
pil_im = Image.open(filename)
open_cv_image = np.array(pil_im)
snapshot_im = pil_im
common.set_input(interpreter, pil_im)
interpreter.invoke()
objs = get_output(interpreter,
score_threshold=args.threshold,
top_k=args.top_k)
#print(objs)
open_cv_image = append_objs_to_img(open_cv_image, objs, labels)
cv2_im_rgb = cv2.cvtColor(open_cv_image, cv2.COLOR_BGR2RGB)
(flag, encodedImage) = cv2.imencode(".jpeg", cv2_im_rgb)
#print(flag)
#print(encodedImage)
f = open("./results/" + dirname + "/" + name, "wb")
f.write(encodedImage)
f.close()
def main():
default_model_dir = '../all_models'
default_model = 'mobilenet_ssd_v2_coco_quant_postprocess_edgetpu.tflite'
default_labels = 'coco_labels.txt'
parser = argparse.ArgumentParser()
parser.add_argument('--model',
help='.tflite model path',
default=os.path.join(default_model_dir, default_model))
parser.add_argument('--labels',
help='label file path',
default=os.path.join(default_model_dir,
default_labels))
parser.add_argument(
'--top_k',
type=int,
default=3,
help='number of categories with highest score to display')
parser.add_argument('--camera_idx',
type=int,
help='Index of which video source to use. ',
default=0)
parser.add_argument('--threshold',
type=float,
default=0.1,
help='classifier score threshold')
args = parser.parse_args()
print('Loading {} with {} labels.'.format(args.model, args.labels))
interpreter = common.make_interpreter(args.model)
interpreter.allocate_tensors()
labels = load_labels(args.labels)
cap = cv2.VideoCapture(args.camera_idx)
while cap.isOpened():
ret, frame = cap.read()
if not ret:
break
cv2_im = frame
cv2_im_rgb = cv2.cvtColor(cv2_im, cv2.COLOR_BGR2RGB)
pil_im = Image.fromarray(cv2_im_rgb)
common.set_input(interpreter, pil_im)
interpreter.invoke()
objs = get_output(interpreter,
score_threshold=args.threshold,
top_k=args.top_k)
cv2_im = append_objs_to_img(cv2_im, objs, labels)
#cv2.imshow('frame', cv2_im)
#cv2_im_rgb = cv2.cvtColor(cv2_im, cv2.COLOR_BGR2RGB)
#pil_im = Image.fromarray(cv2_im_rgb)
#handle_image_conversion(pil_im)
if cv2.waitKey(1) & 0xFF == ord('q'):
break
cap.release()
cv2.destroyAllWindows()
def main():
default_model_dir = "../coral/models"
default_label_dir = "../coral/labels"
default_model = "mobilenet_v2_1.0_224_quant_edgetpu.tflite"
default_labels = "imagenet_labels.txt"
parser = argparse.ArgumentParser()
parser.add_argument(
"--model",
help=".tflite model path",
default=os.path.join(default_model_dir, default_model),
)
parser.add_argument(
"--labels",
help="label file path",
default=os.path.join(default_label_dir, default_labels),
)
args = parser.parse_args()
with open(args.labels, "r") as f:
pairs = (l.strip().split(maxsplit=1) for l in f.readlines())
labels = dict((int(k), v) for k, v in pairs)
interpreter = common.make_interpreter(args.model)
interpreter.allocate_tensors()
with picamera.PiCamera() as camera:
camera.resolution = (640, 480)
camera.framerate = 30
camera.annotate_text_size = 20
width, height, channels = common.input_image_size(interpreter)
camera.start_preview()
try:
stream = io.BytesIO()
fps = deque(maxlen=20)
fps.append(time.time())
for foo in camera.capture_continuous(stream,
format="rgb",
use_video_port=True,
resize=(width, height)):
stream.truncate()
stream.seek(0)
input = np.frombuffer(stream.getvalue(), dtype=np.uint8)
start_ms = time.time()
common.input_tensor(interpreter)[:, :] = np.reshape(
input, common.input_image_size(interpreter))
interpreter.invoke()
results = get_output(interpreter, top_k=3, score_threshold=0)
inference_ms = (time.time() - start_ms) * 1000.0
fps.append(time.time())
fps_ms = len(fps) / (fps[-1] - fps[0])
camera.annotate_text = "Inference: {:5.2f}ms FPS: {:3.1f}".format(
inference_ms, fps_ms)
for result in results:
camera.annotate_text += "\n{:.0f}% {}".format(
100 * result[1], labels[result[0]])
print(camera.annotate_text)
sleep(2)
finally:
camera.stop_preview()
def main():
default_model_dir = 'model'
default_model = 'ssdlite_mobiledet_quant_postprocess_edgetpu.tflite'
parser = argparse.ArgumentParser()
parser.add_argument('--model',
help='.tflite model path',
default=os.path.join(default_model_dir, default_model))
parser.add_argument(
'--top_k',
type=int,
default=3,
help='number of categories with highest score to display')
parser.add_argument('--threshold',
type=float,
default=0.5,
help='classifier score threshold')
parser.add_argument('--videosrc',
help='Which video source to use. ',
default='/dev/video0')
parser.add_argument('--videofmt',
help='Input video format.',
default='raw',
choices=['raw', 'h264', 'jpeg'])
args = parser.parse_args()
interpreter = common.make_interpreter(args.model)
interpreter.allocate_tensors()
w, h, _ = common.input_image_size(interpreter)
inference_size = (w, h)
# Average fps over last 30 frames.
fps_counter = common.avg_fps_counter(100)
def user_callback(input_tensor, src_size, inference_box):
nonlocal fps_counter
start_time = time.monotonic()
common.set_input(interpreter, input_tensor)
interpreter.invoke()
# For larger input image sizes, use the edgetpu.classification.engine for better performance
objs = get_output(interpreter, args.threshold, args.top_k)
end_time = time.monotonic()
text_lines = [
'Inference: {:.2f} ms'.format((end_time - start_time) * 1000),
'FPS: {} fps'.format(round(next(fps_counter))),
]
print(' '.join(text_lines))
return generate_svg(src_size, inference_size, inference_box, objs,
text_lines)
result = gstreamer.run_pipeline(user_callback,
src_size=(640, 480),
appsink_size=inference_size,
videosrc=args.videosrc,
videofmt=args.videofmt)
def main():
default_model_dir = '../all_models'
default_model = 'mobilenet_v2_1.0_224_quant_edgetpu.tflite'
default_labels = 'imagenet_labels.txt'
parser = argparse.ArgumentParser()
parser.add_argument('--model', help='.tflite model path',
default=os.path.join(default_model_dir,default_model))
parser.add_argument('--labels', help='label file path',
default=os.path.join(default_model_dir, default_labels))
parser.add_argument('--top_k', type=int, default=3,
help='number of categories with highest score to display')
parser.add_argument('--threshold', type=float, default=0.1,
help='classifier score threshold')
parser.add_argument('--videosrc', help='Which video source to use. ',
default='/dev/video0')
parser.add_argument('--videofmt', help='Input video format.',
default='raw',
choices=['raw', 'h264', 'jpeg'])
args = parser.parse_args()
print('Loading {} with {} labels.'.format(args.model, args.labels))
interpreter = common.make_interpreter(args.model)
interpreter.allocate_tensors()
labels = load_labels(args.labels)
w, h, _ = common.input_image_size(interpreter)
inference_size = (w, h)
# Average fps over last 30 frames.
fps_counter = common.avg_fps_counter(30)
def user_callback(input_tensor, src_size, inference_box):
nonlocal fps_counter
start_time = time.monotonic()
common.set_input(interpreter, input_tensor)
interpreter.invoke()
# For larger input image sizes, use the edgetpu.classification.engine for better performance
results = get_output(interpreter, args.top_k, args.threshold)
end_time = time.monotonic()
text_lines = [
' ',
'Inference: {:.2f} ms'.format((end_time - start_time) * 1000),
'FPS: {} fps'.format(round(next(fps_counter))),
]
for result in results:
text_lines.append('score={:.2f}: {}'.format(result.score, labels.get(result.id, result.id)))
print(' '.join(text_lines))
return generate_svg(src_size, text_lines)
result = gstreamer.run_pipeline(user_callback,
src_size=(640, 480),
appsink_size=inference_size,
videosrc=args.videosrc,
videofmt=args.videofmt)
def main():
default_model_dir = '../all_models'
default_model = 'mobilenet_ssd_v2_coco_quant_postprocess_edgetpu.tflite'
default_labels = 'coco_labels.txt'
parser = argparse.ArgumentParser()
parser.add_argument('--model', help='.tflite model path',
default=os.path.join(default_model_dir,default_model))
parser.add_argument('--labels', help='label file path',
default=os.path.join(default_model_dir, default_labels))
parser.add_argument('--top_k', type=int, default=3,
help='number of categories with highest score to display')
parser.add_argument('--camera_idx', type=str, help='Index of which video source to use. ', default = 0)
parser.add_argument('--threshold', type=float, default=0.1,
help='classifier score threshold')
args = parser.parse_args()
print('Loading {} with {} labels.'.format(args.model, args.labels))
interpreter = common.make_interpreter(args.model)
interpreter.allocate_tensors()
labels = load_labels(args.labels)
# imagezmq receiver
image_hub = imagezmq.ImageHub(open_port='tcp://147.47.200.65:35556', REQ_REP=False) # REQ_REP=False: use PUB/SUB (non-block)
#cap = cv2.VideoCapture(args.camera_idx)
while True:
# receive from zmq
timestamp, frame = image_hub.recv_image()
dt = datetime.fromtimestamp(timestamp)
#frame = cv2.cvtColor(frame, cv2.COLOR_BGR2RGB)
cv2_im = frame
cv2_im_rgb = cv2.cvtColor(cv2_im, cv2.COLOR_BGR2RGB)
pil_im = Image.fromarray(cv2_im_rgb)
start = time.monotonic()
common.set_input(interpreter, pil_im)
interpreter.invoke()
objs = get_output(interpreter, score_threshold=args.threshold, top_k=args.top_k)
inference_time = time.monotonic() - start
inference_time = 'Inference time: %.2f ms (%.2f fps)' % (inference_time * 1000, 1.0 / inference_time)
cv2_im = append_objs_to_img(cv2_im, objs, labels, inference_time, dt)
#cv2_im = cv2.resize(cv2_im, (720, 720))
cv2.namedWindow("frame", cv2.WND_PROP_FULLSCREEN)
cv2.setWindowProperty("frame",cv2.WND_PROP_FULLSCREEN,cv2.WINDOW_FULLSCREEN)
cv2.imshow("frame", cv2_im)
if cv2.waitKey(1) & 0xFF == ord('q'):
break
cv2.destroyAllWindows()
def main():
default_model_dir = '../all_models'
default_model = 'mobilenet_v2_1.0_224_quant_edgetpu.tflite'
default_labels = 'imagenet_labels.txt'
parser = argparse.ArgumentParser()
parser.add_argument('--model',
help='.tflite model path',
default=os.path.join(default_model_dir, default_model))
parser.add_argument('--labels',
help='label file path',
default=os.path.join(default_model_dir,
default_labels))
args = parser.parse_args()
with open(args.labels, 'r') as f:
pairs = (l.strip().split(maxsplit=1) for l in f.readlines())
labels = dict((int(k), v) for k, v in pairs)
interpreter = common.make_interpreter(args.model)
interpreter.allocate_tensors()
pygame.init()
pygame.camera.init()
camlist = pygame.camera.list_cameras()
print('By default using camera: ', camlist[-1])
camera = pygame.camera.Camera(camlist[-1], (640, 480))
width, height, channels = common.input_image_size(interpreter)
camera.start()
try:
fps = deque(maxlen=20)
fps.append(time.time())
while True:
imagen = camera.get_image()
imagen = pygame.transform.scale(imagen, (width, height))
input = np.frombuffer(imagen.get_buffer(), dtype=np.uint8)
start_ms = time.time()
common.input_tensor(interpreter)[:, :] = np.reshape(
input, (common.input_image_size(interpreter)))
interpreter.invoke()
results = get_output(interpreter, top_k=3, score_threshold=0)
inference_ms = (time.time() - start_ms) * 1000.0
fps.append(time.time())
fps_ms = len(fps) / (fps[-1] - fps[0])
annotate_text = 'Inference: {:5.2f}ms FPS: {:3.1f}'.format(
inference_ms, fps_ms)
for result in results:
annotate_text += '\n{:.0f}% {}'.format(100 * result[1],
labels[result[0]])
print(annotate_text)
finally:
camera.stop()
def detect_object(args):
global outputFrame, lock
interpreter = common.make_interpreter(args.model)
interpreter.allocate_tensors()
labels = load_labels(args.labels)
if args.videosrc=='dev':
cap = cv2.VideoCapture(args.camera_idx)
else:
if args.netsrc==None:
print("--videosrc was set to net but --netsrc was not specified")
sys.exit()
cap = cv2.VideoCapture(args.netsrc)
while cap.isOpened():
ret, frame = cap.read()
if not ret:
break
cv2_im = frame
cv2_im_rgb = cv2.cvtColor(cv2_im, cv2.COLOR_BGR2RGB)
pil_im = Image.fromarray(cv2_im_rgb)
snapshot_im = pil_im
common.set_input(interpreter, pil_im)
interpreter.invoke()
objs = get_output(interpreter, score_threshold=args.threshold, top_k=args.top_k)
cv2_im = append_objs_to_img(cv2_im, objs, labels)
if args.displayBool == 'True':
cv2.imshow('frame', cv2_im)
# acquire the lock, set the output frame, and release the
# lock
with lock:
outputFrame = cv2_im.copy()
if (time.time() - last_save) >=1:
take_snapshot(snapshot_im, objs, labels, exclude=args.exclude.split(','), include=args.include.split(','))
if cv2.waitKey(1) & 0xFF == ord('q'):
break
cap.release()
cv2.destroyAllWindows()
def serve():
default_model_dir = '/media/mendel/detection-server/models'
default_model = 'mobilenet_ssd_v2_coco_quant_postprocess_edgetpu.tflite'
default_labels = 'coco_labels.txt'
parser = argparse.ArgumentParser()
parser.add_argument('--model', help='.tflite model path',
default=os.path.join(default_model_dir,default_model))
parser.add_argument('--labels', help='label file path',
default=os.path.join(default_model_dir, default_labels))
parser.add_argument('--camera_idx', type=int, help='Index of which video source to use. ', default=1)
parser.add_argument('--threshold', type=float, help='Detector threshold. ', default=0.7)
parser.add_argument('--display', dest='display', action='store_true', help='Display object data. ')
parser.set_defaults(display=False)
args = parser.parse_args()
print('Loading {} with {} labels.'.format(args.model, args.labels))
interpreter = common.make_interpreter(os.path.join(default_model_dir, args.model))
interpreter.allocate_tensors()
labels = common.load_labels(os.path.join(default_model_dir, args.labels))
# Get native camera resolution.
cap = cv2.VideoCapture(args.camera_idx)
camera_res = int(cap.get(cv2.CAP_PROP_FRAME_WIDTH)), int(cap.get(cv2.CAP_PROP_FRAME_HEIGHT))
cap.release()
with concurrent.futures.ThreadPoolExecutor(max_workers=10) as executor:
# Start a thread to detect objects in camera frames.
future = executor.submit(start_detector, args.camera_idx, interpreter,
args.threshold, labels, camera_res, args.display)
# Start other threads for the gprc server.
server = grpc.server(executor)
detection_server_pb2_grpc.add_DetectionServerServicer_to_server(
DetectionServerServicer(camera_res), server)
server.add_insecure_port('[::]:50051')
server.start()
# Show the value returned by the executor.submit call.
# This will wait forever unless a runtime error is encountered.
future.result()
server.stop(None)
def __init__(self, configuration):
self.log_directory = configuration['log_directory']
self.model_directory = configuration['model_directory']
self.model = os.path.join(self.model_directory, configuration['model'])
self.labels_file = os.path.join(self.model_directory,
configuration['labels'])
self.top_k = int(configuration['top_k'])
self.camera_id = int(configuration['camera_id'])
self.score_threshold = float(configuration['score_threshold'])
self.clip_duration_sec = int(configuration['clip_duration_sec'])
self.expire_time = self.clip_duration_sec
self.video_directory = configuration['video_directory']
self.camera_stream_url = configuration['camera_stream_url']
print('Loading {} with {} labels.'.format(self.model,
self.labels_file))
self.interpreter = common.make_interpreter(self.model)
self.interpreter.allocate_tensors()
self.stream = cv2.VideoCapture(self.camera_stream_url)
def main():
default_model_dir = '../all_models'
default_model = 'mobilenet_ssd_v2_coco_quant_postprocess_edgetpu.tflite'
default_labels = 'coco_labels.txt'
parser = argparse.ArgumentParser()
parser.add_argument('--model', help='.tflite model path',
default=os.path.join(default_model_dir,default_model))
parser.add_argument('--labels', help='label file path',
default=os.path.join(default_model_dir, default_labels))
parser.add_argument('--top_k', type=int, default=3,
help='number of categories with highest score to display')
parser.add_argument('--threshold', type=float, default=0.1,
help='classifier score threshold')
args = parser.parse_args()
print('Loading {} with {} labels.'.format(args.model, args.labels))
interpreter = common.make_interpreter(args.model)
interpreter.allocate_tensors()
labels = load_labels(args.labels)
w, h, _ = common.input_image_size(interpreter)
inference_size = (w, h)
# Average fps over last 30 frames.
fps_counter = common.avg_fps_counter(30)
def user_callback(input_tensor, src_size, inference_box):
nonlocal fps_counter
start_time = time.monotonic()
common.set_input(interpreter, input_tensor)
interpreter.invoke()
# For larger input image sizes, use the edgetpu.classification.engine for better performance
objs = get_output(interpreter, args.threshold, args.top_k)
end_time = time.monotonic()
text_lines = [
'Inference: {:.2f} ms'.format((end_time - start_time) * 1000),
'FPS: {} fps'.format(round(next(fps_counter))),
]
print(' '.join(text_lines))
return generate_svg(src_size, inference_size, inference_box, objs, labels, text_lines)
result = gstreamer.run_pipeline(user_callback, appsink_size=inference_size)
def classify(model_type=ModelType.General, top_k=1):
interpreter = common.make_interpreter(model_type.model_path())
interpreter.allocate_tensors()
labels = load_labels(model_type.label_path())
cap = cv2.VideoCapture(0)
if cap.isOpened():
for i in range(0,15):
ret, frame = cap.read()
time.sleep(1/1000)
if not ret:
break
cv2_im_rgb = cv2.cvtColor(frame, cv2.COLOR_BGR2RGB)
pil_im = Image.fromarray(cv2_im_rgb)
common.set_input(interpreter, pil_im)
results = classify_image(interpreter, pil_im, top_k)
for label_id, prob in results:
cv2.putText(cv2_im_rgb, labels[label_id], (5,35), cv2.FONT_HERSHEY_SIMPLEX, .7, (0,0,0), 2)
print('%s: %.5f' % (labels[label_id], prob))
cv2.imshow('Classification', cv2_im_rgb)
cv2.waitKey(50)
def make(obj):
fs = "{0}({1})"
parsed = parse.parse(fs, labels[obj[0]])
if parsed != None and len(parsed.fixed) > 1:
tLabel = parsed[1]
else:
tLabel = labels[obj[0]]
return Result(
label = tLabel,
percent = int(100 * obj[1])
)
cap.release()
return [make(obj) for obj in results]
def main():
cam_w, cam_h = 640, 480
default_model_dir = '../all_models'
default_model = 'mobilenet_ssd_v2_coco_quant_postprocess_edgetpu.tflite'
default_labels = 'coco_labels.txt'
parser = argparse.ArgumentParser()
parser.add_argument('--model', help='.tflite model path',
default=os.path.join(default_model_dir,default_model))
parser.add_argument('--labels', help='label file path',
default=os.path.join(default_model_dir, default_labels))
parser.add_argument('--top_k', type=int, default=5,
help='number of categories with highest score to display')
parser.add_argument('--threshold', type=float, default=0.5,
help='classifier score threshold')
args = parser.parse_args()
with open(args.labels, 'r') as f:
pairs = (l.strip().split(maxsplit=1) for l in f.readlines())
labels = dict((int(k), v) for k, v in pairs)
print('Loading {} with {} labels.'.format(args.model, args.labels))
interpreter = common.make_interpreter(args.model)
interpreter.allocate_tensors()
labels = load_labels(args.labels)
pygame.init()
pygame.font.init()
font = pygame.font.SysFont('Arial', 20)
pygame.camera.init()
camlist = pygame.camera.list_cameras()
w, h, _ = common.input_image_size(interpreter)
print('By default using camera: ', camlist[-1])
camera = pygame.camera.Camera(camlist[-1], (cam_w, cam_h))
try:
display = pygame.display.set_mode((cam_w, cam_h), 0)
except pygame.error as e:
sys.stderr.write("\nERROR: Unable to open a display window. Make sure a monitor is attached and that "
"the DISPLAY environment variable is set. Example: \n"
">export DISPLAY=\":0\" \n")
raise e
red = pygame.Color(255, 0, 0)
camera.start()
try:
last_time = time.monotonic()
while True:
mysurface = camera.get_image()
imagen = pygame.transform.scale(mysurface, (w, h))
input = np.frombuffer(imagen.get_buffer(), dtype=np.uint8)
start_time = time.monotonic()
common.input_tensor(interpreter)[:,:] = np.reshape(input, (common.input_image_size(interpreter)))
interpreter.invoke()
results = get_output(interpreter, score_threshold=args.threshold, top_k=args.top_k)
stop_time = time.monotonic()
inference_ms = (stop_time - start_time)*1000.0
fps_ms = 1.0 / (stop_time - last_time)
last_time = stop_time
annotate_text = 'Inference: {:5.2f}ms FPS: {:3.1f}'.format(inference_ms, fps_ms)
for result in results:
x0, y0, x1, y1 = list(result.bbox)
rect = pygame.Rect(x0 * cam_w, y0 * cam_h, (x1 - x0) * cam_w, (y1 - y0) * cam_h)
pygame.draw.rect(mysurface, red, rect, 1)
label = '{:.0f}% {}'.format(100*result.score, labels.get(result.id, result.id))
text = font.render(label, True, red)
print(label, ' ', end='')
mysurface.blit(text, (x0 * cam_w , y0 * cam_h))
text = font.render(annotate_text, True, red)
print(annotate_text)
mysurface.blit(text, (0, 0))
display.blit(mysurface, (0, 0))
pygame.display.flip()
finally:
camera.stop()
def main():
parser = argparse.ArgumentParser()
parser.add_argument('--camera_idx',
type=str,
help='Index of which video source to use. ',
default=0)
parser.add_argument(
'--model',
type=str,
help='Pose model to use. ',
default=
'models/posenet_mobilenet_v1_075_481_641_quant_decoder_edgetpu.tflite')
parser.add_argument('--pose3d',
type=str,
help='3D Pose model to use. ',
default='models/3dpose_gan_edgetpu.tflite')
parser.add_argument('--dataset',
type=str,
help='Type of dataset. ',
default="CORAL")
parser.add_argument('--rot',
type=int,
help='Number of degree to rotate in 3D pose. ',
default=90)
args = parser.parse_args()
engine = PoseEngine(args.model)
_, image_height, image_width, _ = engine.get_input_tensor_shape()
interpreter_3dpose = None
if len(args.pose3d) > 0:
interpreter_3dpose = common.make_interpreter(args.pose3d)
interpreter_3dpose.allocate_tensors()
print("Load all models done!")
cap = cv2.VideoCapture(args.camera_idx)
while cap.isOpened():
ret, frame = cap.read()
if not ret:
break
cv2_im = frame
cv2_im_rgb = cv2.cvtColor(cv2_im, cv2.COLOR_BGR2RGB)
pil_image = Image.fromarray(cv2_im_rgb)
pil_image.resize((image_width, image_height), Image.NEAREST)
start_time = time.monotonic()
poses, inference_time = engine.DetectPosesInImage(np.uint8(pil_image))
print('2D Pose Inference time: %.fms (%.2f fps)' %
(inference_time, 1000 / inference_time))
cv2_im = draw_skel_and_kp(cv2_im, poses, args.rot, interpreter_3dpose)
end_time = time.monotonic()
process_time = 1000 * (end_time - start_time)
print('3D Pose End-to-End Inference time: %.fms (%.2f fps)' %
(process_time, 1000 / process_time))
cv2.namedWindow("frame", cv2.WND_PROP_FULLSCREEN)
cv2.setWindowProperty("frame", cv2.WND_PROP_FULLSCREEN,
cv2.WINDOW_FULLSCREEN)
cv2.imshow('frame', cv2_im)
if cv2.waitKey(1) & 0xFF == ord('q'):
break
cap.release()
cv2.destroyAllWindows()
def main():
default_model_dir = '../all_models'
default_model = 'mobilenet_ssd_v2_face_quant_postprocess_edgetpu.tflite'
default_labels = 'fer_labels.txt'
parser = argparse.ArgumentParser()
parser.add_argument('--model',
help='.tflite model path',
default=os.path.join(default_model_dir, default_model))
parser.add_argument('--labels',
help='label file path',
default=os.path.join(default_model_dir,
default_labels))
parser.add_argument(
'--top_k',
type=int,
default=1,
help='number of categories with highest score to display')
parser.add_argument('--threshold',
type=float,
default=0.1,
help='classifier score threshold')
parser.add_argument('--videosrc',
help='Which video source to use. ',
default='/dev/video0')
parser.add_argument('--videofmt',
help='Input video format.',
default='raw',
choices=['raw', 'h264', 'jpeg'])
args = parser.parse_args()
print('Loading {} with {} labels.'.format(args.model, args.labels))
face_interpreter = common.make_interpreter(
os.path.join(default_model_dir, default_model))
face_interpreter.allocate_tensors()
# fer interpreter
fer_interpreter = common.make_interpreter(args.model)
fer_interpreter.allocate_tensors()
labels = load_labels(args.labels)
w, h, _ = common.input_image_size(face_interpreter)
inference_size = (w, h)
# Average fps over last 30 frames.
fps_counter = common.avg_fps_counter(30)
def user_callback(input_tensor, src_size, inference_box):
nonlocal fps_counter
start_time = time.monotonic()
common.set_input(face_interpreter, input_tensor)
face_interpreter.invoke()
# For larger input image sizes, use the edgetpu.classification.engine for better performance
objs = get_output(face_interpreter, args.threshold, args.top_k)
# Get face detected part
from PIL import Image
im = Image.fromarray(common.input_tensor(face_interpreter))
src_w, src_h = src_size
inf_w, inf_h = inference_size
results = []
emo_objs = []
for obj in objs:
x0, y0, x1, y1 = list(obj.bbox)
# Relative coordinates.
x, y, w, h = x0, y0, x1 - x0, y1 - y0
# Absolute coordinates, input tensor space.
x, y, w, h = int(x * inf_w), int(y * inf_h), int(w * inf_w), int(
h * inf_h)
crop_rectangle = (x, y, x + w, y + h)
# get face
face = im.crop(crop_rectangle)
face = np.array(face)
# convert to grayscale
#face = cv2.cvtColor(face, cv2.COLOR_RGB2GRAY)
print(face.shape)
face = cv2.resize(face, (224, 224))
face = face.astype(np.uint8)
#face /= float(face.max())
face = np.reshape(face.flatten(), (224, 224, 3))
# invoke fer interpreter
common.set_input2(fer_interpreter, face)
fer_interpreter.invoke()
# process results
results = get_emotion(fer_interpreter)
if len(results) > 0:
setattr(obj, "id", results[0].id)
setattr(obj, "score", results[0].score)
emo_objs.append(obj)
objs = emo_objs
end_time = time.monotonic()
text_lines = []
if len(objs) > 0:
text_lines = [
'Inference: {:.2f} ms'.format((end_time - start_time) * 1000),
'FPS: {} fps'.format(round(next(fps_counter))),
]
for result in results:
text_lines.append('score={:.2f}: {}'.format(
result.score, labels.get(result.id, result.id)))
#print(' '.join(text_lines))
return generate_svg(src_size, inference_size, inference_box, objs,
labels, text_lines)
result = gstreamer.run_pipeline(user_callback,
src_size=(640, 480),
appsink_size=inference_size,
videosrc=args.videosrc,
videofmt=args.videofmt)
def main():
default_model_dir = '../all_models'
default_model = 'mobilenet_ssd_v2_face_quant_postprocess_edgetpu.tflite'
default_labels = 'coco_labels.txt'
parser = argparse.ArgumentParser()
parser.add_argument('--model',
help='.tflite model path',
default=os.path.join(default_model_dir, default_model))
parser.add_argument('--labels',
help='label file path',
default=os.path.join(default_model_dir,
default_labels))
parser.add_argument(
'--top_k',
type=int,
default=3,
help='number of categories with highest score to display')
parser.add_argument('--threshold',
type=float,
default=0.1,
help='classifier score threshold')
args = parser.parse_args()
print('Loading {} with {} labels.'.format(args.model, args.labels))
interpreter = common.make_interpreter(args.model)
interpreter.allocate_tensors()
labels = load_labels(args.labels)
# csv writer
f = open('face_output.csv', 'w')
with f:
fnames = [
'timestamp', 'idx', 'label', 'width', 'height', 'xmin', 'ymin',
'xmax', 'ymax', 'score'
]
writer = csv.DictWriter(f, fieldnames=fnames)
writer.writeheader()
# read frames
for image_path in sorted(
glob.glob('/home/mendel/dataset/Store/frames/Camera01/*.jpg')):
image_name = os.path.splitext(os.path.basename(image_path))[0]
#print(image_name)
pil_im = Image.open(image_path)
common.set_input(interpreter, pil_im)
interpreter.invoke()
objs = get_output(interpreter,
score_threshold=args.threshold,
top_k=args.top_k)
(width, height) = pil_im.size
idx = -1
for obj in objs:
x0, y0, x1, y1 = list(obj.bbox)
x0, y0, x1, y1 = int(x0 * width), int(y0 * height), int(
x1 * width), int(y1 * height)
score = obj.score
label = 'face'
idx += 1
writer.writerow({
'timestamp': image_name,
'idx': idx,
'label': label,
'width': width,
'height': height,
'xmin': x0,
'ymin': y0,
'xmax': x1,
'ymax': y1,
'score': score
})
'bitrate': 1000000,
'source': '/dev/video2:YUY2:640x480:30/1',
'window': 10,
'top_k': 3,
'threshold': 0.1,
'print': False,
'camera_idx': 2,
'labels': os.path.join(default_model_dir, default_labels)
})
#Start up camera
cap = open_available_stream()
#Load up model, labels, and interpreter
print('Loading {} with {} labels.'.format(args.model, args.labels))
interpreter = common.make_interpreter(args.model)
interpreter.allocate_tensors()
labels = load_labels(args.labels)
#Start networktables
NetworkTables.initialize()
sd = NetworkTables.getTable("SmartDashboard2")
def main():
#Run our flask app
app.run(host='0.0.0.0', port=5000, use_reloader=False)
#Cleaning up
cap.release()
def main():
default_model_dir = '../all_models'
default_model = 'mobilenet_ssd_v2_coco_quant_postprocess_edgetpu.tflite'
parser = argparse.ArgumentParser()
parser.add_argument('--model',
help='.tflite model path',
default=os.path.join(default_model_dir, default_model))
parser.add_argument(
'--top_k',
type=int,
default=5,
help='number of categories with highest score to display')
parser.add_argument('--camera_idx',
type=int,
help='Index of which video source to use. ',
default=0)
parser.add_argument('--threshold',
type=float,
default=0.5,
help='classifier score threshold')
parser.add_argument('--video_width',
type=int,
help='Width resolution of the Video Capture',
default=1920)
parser.add_argument('--video_height',
type=int,
help='Width resolution of the Video Capture',
default=1080)
parser.add_argument(
'--confirmations',
type=int,
help=
'Frames detected with one or more person(s) needed before sending out an alert',
default=30)
parser.add_argument(
'--time_period',
type=int,
help='Maximum time for confirmation check (in seconds)',
default=10)
parser.add_argument('--alert_cooldown',
type=int,
help='Cooldown time between alerts (in seconds)',
default=600)
args = parser.parse_args()
print('Loading {}.'.format(args.model))
interpreter = common.make_interpreter(args.model)
interpreter.allocate_tensors()
cap = cv2.VideoCapture(args.camera_idx)
# set VideoCapture resolution
cap.set(cv2.CAP_PROP_FRAME_WIDTH, args.video_width)
cap.set(cv2.CAP_PROP_FRAME_HEIGHT, args.video_height)
cooldown_until = 0
confirmations = array.array('d', [0] * args.confirmations)
while cap.isOpened():
ret, frame = cap.read()
if not ret:
break
cv2_im = frame
cv2_im_rgb = cv2.cvtColor(cv2_im, cv2.COLOR_BGR2RGB)
pil_im = Image.fromarray(cv2_im_rgb)
common.set_input(interpreter, pil_im)
interpreter.invoke()
objs = get_output(interpreter,
score_threshold=args.threshold,
top_k=args.top_k)
persons = list(filter(lambda x: x.id == PERSON_COCO_INDEX, objs))
if persons:
current_time = time.time()
confirmations.append(current_time)
if confirmations[-1] - confirmations.pop(
0
) <= args.time_period and confirmations[-1] >= cooldown_until:
print("alerted at", current_time)
image_link = upload_image(pil_im)
send_alert("person detected at {}. \n{}".format(
datetime.datetime.now().isoformat(), image_link))
cooldown_until = current_time + args.alert_cooldown
if cv2.waitKey(1) & 0xFF == ord('q'):
break
cap.release()
cv2.destroyAllWindows()
def main():
default_model_dir = '../all_models'
default_model = 'mobilenet_ssd_v2_coco_quant_postprocess_edgetpu.tflite'
default_labels = 'coco_labels.txt'
parser = argparse.ArgumentParser()
parser.add_argument('--model',
help='.tflite model path',
default=os.path.join(default_model_dir, default_model))
parser.add_argument('--labels',
help='label file path',
default=os.path.join(default_model_dir,
default_labels))
parser.add_argument(
'--top_k',
type=int,
default=3,
help='number of categories with highest score to display')
parser.add_argument('--threshold',
type=float,
default=0.1,
help='classifier score threshold')
args = parser.parse_args()
print('Loading {} with {} labels.'.format(args.model, args.labels))
interpreter = common.make_interpreter(args.model)
interpreter.allocate_tensors()
labels = load_labels(args.labels)
cap = cv2.VideoCapture(0)
###############################For sending to the socket
#cap.set(3, 320);
#cap.set(4, 240);
img_counter = 0
encode_param = [int(cv2.IMWRITE_JPEG_QUALITY), 90]
######################################################################33
while cap.isOpened():
ret, frame = cap.read()
if not ret:
break
cv2_im = frame
cv2_im_rgb = cv2.cvtColor(cv2_im, cv2.COLOR_BGR2RGB)
pil_im = Image.fromarray(cv2_im_rgb)
common.set_input(interpreter, pil_im)
interpreter.invoke()
objs = get_output(interpreter,
score_threshold=args.threshold,
top_k=args.top_k)
cv2_im = append_objs_to_img(cv2_im, objs, labels)
#################SENDING TO THE SERVER#################3
result, frame = cv2.imencode('.jpg', cv2_im, encode_param)
data = pickle.dumps(frame, 0)
size = len(data)
clientsocket.sendall(struct.pack(">L", size) + data)
#connection.sendall(struct.pack(">L", size) + data)
img_counter += 1
#################################################################
#cv2.imshow('frame', cv2_im)
#if cv2.waitKey(1) & 0xFF == ord('q'):
# break
cap.release()
cv2.destroyAllWindows()
def main():
default_model_dir = '../all_models'
default_model = 'mobilenet_ssd_v2_coco_quant_postprocess_edgetpu.tflite'
default_labels = 'coco_labels.txt'
parser = argparse.ArgumentParser()
parser.add_argument('--model',
help='.tflite model path',
default=os.path.join(default_model_dir, default_model))
parser.add_argument('--labels',
help='label file path',
default=os.path.join(default_model_dir,
default_labels))
parser.add_argument(
'--top_k',
type=int,
default=10,
help='number of categories with highest score to display')
parser.add_argument('--threshold',
type=float,
default=0.3,
help='classifier score threshold')
parser.add_argument('--class_ids',
nargs='*',
type=int,
default=0,
help='Array of class id')
parser.add_argument('--input_files',
default='/home/mendel/dataset/*.jpg',
help='Input files')
parser.add_argument('--csv_out',
default='detect_output.csv',
help='csv output file')
args = parser.parse_args()
if args.class_ids == 0:
args.class_ids = [0]
print('Loading {} with {} labels.'.format(args.model, args.labels))
interpreter = common.make_interpreter(args.model)
interpreter.allocate_tensors()
labels = load_labels(args.labels)
# csv writer
f = open(args.csv_out, 'w')
with f:
fnames = [
'timestamp', 'idx', 'label', 'width', 'height', 'xmin', 'ymin',
'xmax', 'ymax', 'score'
]
writer = csv.DictWriter(f, fieldnames=fnames)
writer.writeheader()
# read frames
inference_time = []
for image_path in sorted(glob.glob(args.input_files)):
image_name = os.path.splitext(os.path.basename(image_path))[0]
#print(image_name)
pil_im = Image.open(image_path)
# inference
start = time.time()
common.set_input(interpreter, pil_im)
interpreter.invoke()
objs = get_output(interpreter,
score_threshold=args.threshold,
top_k=args.top_k,
class_list=args.class_ids)
inference_time.append(time.time() - start)
# return results
(width, height) = pil_im.size
idx = -1
for obj in objs:
x0, y0, x1, y1 = list(obj.bbox)
x0, y0, x1, y1 = int(x0 * width), int(y0 * height), int(
x1 * width), int(y1 * height)
score = obj.score
label = labels.get(obj.id, obj.id)
idx += 1
writer.writerow({
'timestamp': image_name,
'idx': idx,
'label': label,
'width': width,
'height': height,
'xmin': x0,
'ymin': y0,
'xmax': x1,
'ymax': y1,
'score': score
})
print("Inference time : {:.3f} ms".format(
sum(inference_time) * 1000 / len(inference_time)))
print("Frames per second : {:.2f} fps".format(
len(inference_time) / sum(inference_time)))
def main():
# default_model_dir = '../all_models'
# default_model = 'mobilenet_ssd_v2_coco_quant_postprocess_edgetpu.tflite'
# default_labels = 'coco_labels.txt'
default_model_dir = '../cmpe297_model'
# default_model = 'ssdlite_6C_SB_10K_mobiledet_screws.tflite' #5 classes small BB
# default_model = 'ssdlite_6C_SB_10K_mobiledet_screws_edgetpu.tflite' #5 classes small BB
# default_model = 'ssdlite_6C_SB_25K_mobiledet_screws.tflite' #5 classes small BB
default_model = 'ssdlite_6C_SB_25K_mobiledet_screws_edgetpu.tflite' #5 classes small BB
# default_model = 'ssdlite_6C_BB_10K_mobiledet_screws.tflite' #5 classes big BB 1K
# default_model = 'ssdlite_6C_BB_10K_mobiledet_screws_edgetpu.tflite' #5 classes big BB 1K
default_labels = 'ssdlite_mobiledet_screws_6c_labels.txt'
# default_model = 'ssdlite_2C_BB_10K_mobiledet_screws.tflite' #5 classes big BB 1K
# default_model = 'ssdlite_2C_BB_10K_mobiledet_screws_edgetpu.tflite' #5 classes big BB 1K
# default_labels = 'ssdlite_mobiledet_screws_2c_labels.txt'
# default_model_dir = '../cmpe297_model'
# default_model = 'Sergio_v3_ssdlite_mobiledet_dog_vs_cat.tflite'
# # default_model = 'Sergio_v3_sdlite_mobiledet_dog_vs_cat_edgetpu.tflite'
# default_labels = 'cat_vs_doc_All.txt'
# default_model = 'mobilenet_v2_1.0_224_quant_edgetpu_cmpe297.tflite'
# # default_model = 'mobilenet_v2_1.0_224_quant_cmpe297.tflite'
# default_labels = 'flower_labels_cmpe297.txt'
# default_model = 'eager_mobilenet_v2_1.0_224_quant.tflite' #no edgeTPU
# default_model = 'eager_mobilenet_v2_1.0_224_quant_edgetpu.tflite' #eager
#
# default_model = 'eager2_mobilenet_v2_1.0_224_quant.tflite' #eager
# default_model = 'eager2_mobilenet_v2_1.0_224_quant_edgetpu.tflite' #eager
# default_labels = 'duckylabels.txt'
# default_model = 'quant_coco-tiny-v3-relu.tflite'
# default_model = 'quant_coco-tiny-v3-relu_edgetpu.tflite'
# default_model = 'ssdlite_mobiledet_dog_vs_cat_edgetpu.tflite'
# default_labels = 'cat_vs_doc.txt'
# default_model = 'cmpe297_ssdlite_mobiledet_dog.tflite'
# default_model = 'cmpe297_ssdlite_mobiledet_dog_edgetpu.tflite'
# default_model = 'cmpe297v2_ssdlite_mobiledet_dog_edgetpu.tflite'
# default_labels = 'dogs_labels.txt'
# default_model = 'ssdlite_mobiledet_dog_vs_cat_edgetpuAcha.tflite'
# default_labels = 'cat_vs_doc_All.txt'
parser = argparse.ArgumentParser()
parser.add_argument('--model',
help='.tflite model path',
default=os.path.join(default_model_dir, default_model))
parser.add_argument('--labels',
help='label file path',
default=os.path.join(default_model_dir,
default_labels))
parser.add_argument(
'--top_k',
type=int,
default=3,
help='number of categories with highest score to display')
parser.add_argument('--threshold',
type=float,
default=0.1,
help='classifier score threshold')
parser.add_argument('--videosrc',
help='Which video source to use. ',
default='/dev/video0')
parser.add_argument('--videofmt',
help='Input video format.',
default='raw',
choices=['raw', 'h264', 'jpeg'])
args = parser.parse_args()
print('Loading {} with {} labels.'.format(args.model, args.labels))
interpreter = common.make_interpreter(args.model)
interpreter.allocate_tensors()
labels = load_labels(args.labels)
w, h, _ = common.input_image_size(interpreter)
inference_size = (w, h)
# Average fps over last 30 frames.
fps_counter = common.avg_fps_counter(30)
def user_callback(input_tensor, src_size, inference_box):
nonlocal fps_counter
start_time = time.monotonic()
common.set_input(interpreter, input_tensor)
interpreter.invoke()
# For larger input image sizes, use the edgetpu.classification.engine for better performance
objs = get_output(interpreter, args.threshold, args.top_k)
# print(objs[0].bbox)
end_time = time.monotonic()
text_lines = [
'Inference: {:.2f} ms'.format((end_time - start_time) * 1000),
'FPS: {} fps'.format(round(next(fps_counter))),
]
print(' '.join(text_lines))
return generate_svg(src_size, inference_size, inference_box, objs,
labels, text_lines)
result = gstreamer.run_pipeline(user_callback,
src_size=(640, 480),
appsink_size=inference_size,
videosrc=args.videosrc,
videofmt=args.videofmt)
def main():
default_model_dir = 'all_models'
default_model = 'mobilenet_ssd_v2_coco_quant_postprocess_edgetpu.tflite'
default_labels = 'coco_labels.txt'
parser = argparse.ArgumentParser()
parser.add_argument('--model',
help='.tflite model path',
default=os.path.join(default_model_dir, default_model))
parser.add_argument('--labels',
help='label file path',
default=os.path.join(default_model_dir,
default_labels))
parser.add_argument(
'--top_k',
type=int,
default=3,
help='number of categories with highest score to display')
parser.add_argument('--camera_idx',
type=int,
help='Index of which video source to use. ',
default=0)
parser.add_argument('--threshold',
type=float,
default=0.1,
help='classifier score threshold')
args = parser.parse_args()
multiTracker = cv2.MultiTracker_create()
#Initialize logging files
logging.basicConfig(filename='storage/results.log',
format='%(asctime)s-%(message)s',
level=logging.DEBUG)
print('Loading {} with {} labels.'.format(args.model, args.labels))
interpreter = common.make_interpreter(args.model)
interpreter.allocate_tensors()
labels = load_labels(args.labels)
vs = PiVideoStream(resolution=(2048, 1536), framerate=32).start()
#cap = cv2.VideoCapture(args.camera_idx)
cap = vs.stream
#cap.set(3, 1920)
#cap.set(4, 1440)
# 4:3 resolutions
# 640×480, 800×600, 960×720, 1024×768, 1280×960, 1400×1050,
# 1440×1080 , 1600×1200, 1856×1392, 1920×1440, 2048×1536
# 5 MP
#cap.set(3, 2048)
#cap.set(4, 1536)
bboxes = []
colors = []
visitation = []
trackers = []
started_tracking = None
last_tracked = None
visitation_id = None
save_one_with_boxes = False
recording = False
out = None
fps = FPS().start()
is_stopped = False
current_fps = 4.0
while vs is not None:
try:
frame = vs.read()
if frame is not None:
if fps._numFrames < 500:
fps.update()
else:
fps.stop()
current_fps = fps.fps()
logging.info("[INFO] elasped time: {:.2f}".format(
fps.elapsed()))
logging.info("[INFO] approx. FPS: {:.2f}".format(
fps.fps()))
fps = FPS().start()
success, boxes = multiTracker.update(frame)
if success:
last_tracked = time.time()
if len(boxes) > 0:
logging.info("success {}".format(success))
logging.info("boxes {}".format(boxes))
cv2_im = frame
#cv2_im_rgb = cv2.cvtColor(cv2_im, cv2.COLOR_BGR2RGB)
pil_im = Image.fromarray(cv2_im)
common.set_input(interpreter, pil_im)
interpreter.invoke()
objs = get_output(interpreter,
score_threshold=args.threshold,
top_k=args.top_k)
height, width, channels = cv2_im.shape
bird_detected = False
boxes_to_draw = []
for obj in objs:
x0, y0, x1, y1 = list(obj.bbox)
x0, y0, x1, y1 = int(x0 * width), int(y0 * height), int(
x1 * width), int(y1 * height)
percent = int(100 * obj.score)
object_label = labels.get(obj.id, obj.id)
label = '{}% {}'.format(percent, object_label)
hdd = psutil.disk_usage('/')
if object_label == 'bird' and percent > 40:
bird_detected = True
new_bird = True
for bbox in boxes:
if intersects(bbox, obj.bbox):
logging.info("intersected.. same bird")
new_bird = False
if new_bird and len(bboxes) == 0:
logging.info("found a new bird")
visitation_id = uuid.uuid4()
started_tracking = time.time()
recording = True
save_one_with_boxes = True
bboxes.append(obj.bbox)
colors.append((randint(64, 255), randint(64, 255),
randint(64, 255)))
tracker = cv2.TrackerCSRT_create()
trackers.append(tracker)
multiTracker.add(tracker, cv2_im, obj.bbox)
if hdd.percent < 95:
boxed_image_path = "storage/detected/boxed_{}_{}_{}.png".format(
time.strftime("%Y-%m-%d_%H-%M-%S"), percent,
visitation_id)
full_image_path = "storage/detected/full_{}_{}_{}.png".format(
time.strftime("%Y-%m-%d_%H-%M-%S"), percent,
visitation_id)
cv2.imwrite(boxed_image_path, cv2_im[y0:y1, x0:x1])
if percent > 95:
cv2.imwrite(full_image_path, cv2_im)
else:
logging.info("Not enough disk space")
percent = int(100 * obj.score)
object_label = labels.get(obj.id, obj.id)
label = '{}% {}'.format(percent, object_label)
# postpone drawing so we don't get lines in the photos
boxes_to_draw.append({
"p1": (x0, y0),
"p2": (x1, y1),
"label": label,
"label_p": (x0, y0 + 30)
})
for box in boxes_to_draw:
cv2_im = cv2.rectangle(cv2_im, box["p1"], box["p2"],
(0, 255, 0), 2)
cv2_im = cv2.putText(cv2_im, box["label"], box["label_p"],
cv2.FONT_HERSHEY_SIMPLEX, 1.0,
(255, 0, 0), 2)
if recording == True:
if out == None:
fourcc = cv2.VideoWriter_fourcc(*'X264')
out = cv2.VideoWriter(
"storage/video/{}.avi".format(visitation_id),
fourcc, 4.0, (2048, 1536))
out.write(cv2_im)
if bird_detected == True and save_one_with_boxes == True:
with_boxes_image_path = "storage/with_boxes/full_{}_{}.png".format(
time.strftime("%Y-%m-%d_%H-%M-%S"), visitation_id)
cv2.imwrite(with_boxes_image_path, cv2_im)
save_one_with_boxes = False
if bird_detected == False and len(trackers) > 0:
now = time.time()
if now - last_tracked > 60:
logging.info("visitation {} lasted {} seconds".format(
visitation_id, now - started_tracking))
logging.info("clearing trackers")
for tracker in trackers:
tracker.clear()
multiTracker = cv2.MultiTracker_create()
boxes = []
colors = []
trackers = []
bboxes = []
recording = False
out.release()
out = None
for i, newbox in enumerate(boxes):
x0, y0, x1, y1 = list(newbox)
x0, y0, x1, y1 = int(x0 * width), int(y0 * height), int(
x1 * width), int(y1 * height)
cv2_im = cv2.rectangle(cv2_im, (x0, y0), (x1, y1),
(0, 0, 255), 2)
cv2.namedWindow('Leroy', cv2.WINDOW_NORMAL)
cv2.resizeWindow('Leroy', 800, 600)
cv2.imshow('Leroy', cv2_im)
except KeyboardInterrupt:
print('Interrupted')
try:
sys.exit(0)
except SystemExit:
os._exit(0)
except:
logging.exception('Something happened.')
if cv2.waitKey(1) & 0xFF == ord('q'):
break
cap.release()
cv2.destroyAllWindows()
out.release()
vs.stop()
def main():
default_model_dir = "../coral/models"
default_label_dir = "../coral/labels"
default_model = "ssd_mobilenet_v2_coco_quant_postprocess_edgetpu.tflite"
default_labels = "coco_labels.txt"
parser = argparse.ArgumentParser()
parser.add_argument(
"--model",
help=".tflite model path",
default=os.path.join(default_model_dir, default_model),
)
parser.add_argument(
"--labels",
help="label file path",
default=os.path.join(default_label_dir, default_labels),
)
parser.add_argument(
"--top_k",
type=int,
default=3,
help="number of categories with highest score to display",
)
parser.add_argument("--camera_idx",
type=int,
help="Index of which video source to use. ",
default=0)
parser.add_argument("--threshold",
type=float,
default=0.1,
help="classifier score threshold")
args = parser.parse_args()
print("Loading {} with {} labels.".format(args.model, args.labels))
interpreter = common.make_interpreter(args.model)
interpreter.allocate_tensors()
labels = load_labels(args.labels)
cap = cv2.VideoCapture(args.camera_idx)
count = 0
while cap.isOpened():
ret, frame = cap.read()
count += 1
if not ret:
break
if count == 20:
break
cv2_im = frame
cv2_im_rgb = cv2.cvtColor(cv2_im, cv2.COLOR_BGR2RGB)
pil_im = Image.fromarray(cv2_im_rgb)
common.set_input(interpreter, pil_im)
interpreter.invoke()
objs = get_output(interpreter,
score_threshold=args.threshold,
top_k=args.top_k)
cv2_im = append_objs_to_img(cv2_im, objs, labels)
# cv2.imshow("frame", cv2_im)
# if cv2.waitKey(1) & 0xFF == ord("q"):
# break
# Saving the image
cv2.imwrite("detectImage.jpg", cv2_im)
cap.release()
cv2.destroyAllWindows()
def main():
args = get_arguments()
MODEL = args.model
LABELS = args.labels
TOP_K = args.top_k
THRESHOLD = args.threshold
CAPTURE_MODE = args.capture_mode
INPUT_VIDEO_DIR = args.input_video_dir
INPUT_VIDEO = str(os.path.join(INPUT_VIDEO_DIR, args.input_video))
OUTPUT_VIDEO_DIR = args.output_video_dir
OUTPUT_VIDEO = str(os.path.join(OUTPUT_VIDEO_DIR, args.output_video))
OUTPUT_VIDEO_RES = args.output_video_res
OUTPUT_VIDEO_FPS = args.output_video_fps
OUTPUT_VIDEO_TIME = args.output_video_time
RECORD_MODE = args.record_mode
MONITOR = args.monitor
DEBUG_MODE = args.debug_mode
# Define the hardware
ui = get_ui()
# Create an interpreter engine based on the tflite model path given
debug_print("Loading {} with {} labels.".format(MODEL, LABELS))
interpreter = common.make_interpreter(MODEL)
interpreter.allocate_tensors()
cap = None
out = None
# Define camera or input video properties
cap = load_cap(CAPTURE_MODE, INPUT_VIDEO)
cap_fps = cap.get(cv2.CAP_PROP_FPS)
cap_width = cap.get(cv2.CAP_PROP_FRAME_WIDTH)
cap_height = cap.get(cv2.CAP_PROP_FRAME_HEIGHT)
# Define output file if recording
out = load_out(cap, CAPTURE_MODE, OUTPUT_VIDEO, OUTPUT_VIDEO_RES, cap_fps, OUTPUT_VIDEO_FPS)
labels = load_labels(LABELS)
frame_count = 0
# Read frame of the capture and feed it to the interpreter. Feed out an editted frame with detected objects encapuslated by bounding box.
# Based on the object(s) captured, replay bluetooth message to the ardunio that controls the stepper motor and animatronics.
while cap.isOpened():
ret, frame = cap.read()
if not ret:
break
objs = []
cv2_im, objs = detect(frame, interpreter, labels, THRESHOLD, TOP_K)
display_frame(cv2_im)
relay_bluetooth_detection(ui, objs, labels, cap_width, cap_height)
if RECORD_MODE:
out.write(cv2_im)
if DEBUG_MODE:
frame_count += 1
seconds = (frame_count/cap_fps)%60
debug_print(seconds)
if (cv2.waitKey(1) & 0xFF == ord('q')) or ((RECORD_MODE or DEBUG_MODE) and seconds > OUTPUT_VIDEO_TIME):
debug_print("Ending capture")
break
cleanup(cap, out)
print("Successful termination")
def main():
default_model_dir = './' #'../all_models'
default_model = 'mobilenet_ssd_v2_coco_quant_postprocess_edgetpu.tflite'
default_labels = 'coco_labels.txt'
parser = argparse.ArgumentParser()
parser.add_argument('--model',
help='.tflite model path',
default=os.path.join(default_model_dir, default_model))
parser.add_argument('--labels',
help='label file path',
default=os.path.join(default_model_dir,
default_labels))
parser.add_argument(
'--top_k',
type=int,
default=3,
help='number of categories with highest score to display')
parser.add_argument('--camera_idx',
help='Index of which video source to use. ',
default=0)
parser.add_argument('--threshold',
type=float,
default=0.5,
help='classifier score threshold')
parser.add_argument('--min_angle',
type=float,
default=0.0,
help='minimum angle for sweep')
parser.add_argument('--max_angle',
type=float,
default=180.0,
help='maximum angle for sweep')
args = parser.parse_args()
print('Initializing servo')
servo, servoData = ServoControls.init(
0,
sweepStepTracking=0.1,
minAngle=args.min_angle,
maxAngle=args.max_angle) #SERVO_PIN, 50)
servoThread = _thread.start_new_thread(control_servo, (servo, servoData))
print('Loading {} with {} labels.'.format(args.model, args.labels))
interpreter = common.make_interpreter(args.model)
interpreter.allocate_tensors()
labels = load_labels(args.labels)
# cv2.VideoCapture.set(v_c, cv2.CAP_PROP_FPS, 15)
cap = cv2.VideoCapture(args.camera_idx)
# Read first frame to get window frame shape
_, frame = cap.read()
if frame is None:
raise Exception('Image not found!')
frameH, frameW, frameChannels = frame.shape
lastTargetLost = None
lastTargetLostTime = datetime.datetime.now()
targetState = TargetState.UNKNOWN
play_sound('searching.mp3')
while cap.isOpened():
ret, frame = cap.read()
if not ret:
break
#frame = cv2.flip(frame, flipCode=1)
frame = imutils.rotate(frame, 90)
h, w, layers = frame.shape
aspect_ratio = w / h
cv2_im = frame # cv2.resize(frame, (1920, 1080))
cv2_im_rgb = cv2.cvtColor(cv2_im, cv2.COLOR_BGR2RGB)
pil_im = Image.fromarray(cv2_im_rgb)
common.set_input(interpreter, pil_im)
interpreter.invoke()
objs = get_output(interpreter,
score_threshold=args.threshold,
top_k=args.top_k)
cv2_im = append_objs_to_img(cv2_im, objs, labels)
face = next(filter(lambda a: a.id == 0, objs), None)
if face != None:
if targetState == TargetState.UNKNOWN:
targetState = TargetState.ACQUIRED
height, width, channels = cv2_im.shape
x0, y0, x1, y1 = list(face.bbox)
x0, y0, x1, y1 = int(x0 * width), int(y0 * height), int(
x1 * width), int(y1 * height)
lastTargetLost = 0
servoData['targetCoordinates'] = [
round(abs((x0 + (x1)) / 2)),
round(abs((y0 + (y1)) / 2))
]
else:
# target may have been lost
if targetState == TargetState.TRACKING:
targetState = TargetState.LOST
# track lost time
lastTargetLostTime = datetime.datetime.now()
if targetState == TargetState.LOST and (
lastTargetLostTime == None or
(datetime.datetime.now() - lastTargetLostTime).seconds > 2):
# if lost for over a second, reset targetState back to default
servoData['targetCoordinates'] = [-1, -1]
play_sound('are-still-there.mp3')
targetState = TargetState.UNKNOWN
lastTargetLostTime = None
if targetState == TargetState.ACQUIRED:
play_sound(TARGET_ACQUIRED)
targetState = TargetState.TRACKING
cv2.imshow('frame', cv2_im)
if cv2.waitKey(1) & 0xFF == ord('q'):
break
RUNNING = False
cap.release()
cv2.destroyAllWindows()
def main():
args = get_arguments()
MODEL = args.model
LABELS = args.labels
TOP_K = args.top_k
THRESHOLD = args.threshold
CAPTURE_MODE = args.capture_mode
INPUT_VIDEO_DIR = args.input_video_dir
INPUT_VIDEO = str(os.path.join(INPUT_VIDEO_DIR, args.input_video))
OUTPUT_VIDEO_DIR = args.output_video_dir
OUTPUT_VIDEO = str(os.path.join(OUTPUT_VIDEO_DIR, args.output_video))
OUTPUT_VIDEO_RES = args.output_video_res
OUTPUT_VIDEO_FPS = args.output_video_fps
OUTPUT_VIDEO_TIME = args.output_video_time
RECORD_MODE = args.record_mode
MONITOR = args.monitor
DEBUG_MODE = args.debug_mode
ui = get_ui()
debug_print("Loading {} with {} labels.".format(MODEL, LABELS))
interpreter = common.make_interpreter(MODEL)
interpreter.allocate_tensors()
cap = None
out = None
cap = load_cap(CAPTURE_MODE, INPUT_VIDEO)
cap_fps = cap.get(cv2.CAP_PROP_FPS)
cap_width = cap.get(cv2.CAP_PROP_FRAME_WIDTH)
cap_height = cap.get(cv2.CAP_PROP_FRAME_HEIGHT)
out = load_out(cap, CAPTURE_MODE, OUTPUT_VIDEO, OUTPUT_VIDEO_RES, cap_fps,
OUTPUT_VIDEO_FPS)
labels = load_labels(LABELS)
frame_count = 0
while cap.isOpened():
ret, frame = cap.read()
if not ret:
break
objs = []
cv2_im, objs = detect(frame, interpreter, labels, THRESHOLD, TOP_K)
display_frame(cv2_im)
relay_bluetooth_detection(ui, objs, cap_width, cap_height)
if RECORD_MODE:
out.write(cv2_im)
frame_count += 1
seconds = (frame_count / cap_fps) % 60
debug_print(seconds)
if (cv2.waitKey(1) & 0xFF
== ord('q')) or (RECORD_MODE and seconds > OUTPUT_VIDEO_TIME):
debug_print("Ending capture")
break
cleanup(cap, out)
print("Successful termination")
def main():
default_model_dir = '../all_models'
default_model = 'mobilenet_ssd_v2_coco_quant_postprocess_edgetpu.tflite'
default_labels = 'coco_labels.txt'
parser = argparse.ArgumentParser()
parser.add_argument('--model', help='.tflite model path',
default=os.path.join(default_model_dir,default_model))
parser.add_argument('--labels', help='label file path',
default=os.path.join(default_model_dir, default_labels))
parser.add_argument('--top_k', type=int, default=3,
help='number of categories with highest score to display')
parser.add_argument('--camera_idx', type=int, help='Index of which video source to use. ', default = 0)
parser.add_argument('--threshold', type=float, default=0.1,
help='classifier score threshold')
args = parser.parse_args()
print('Loading {} with {} labels.'.format(args.model, args.labels))
interpreter = common.make_interpreter(args.model)
interpreter.allocate_tensors()
labels = load_labels(args.labels)
cap = cv2.VideoCapture(1)
while cap.isOpened():
ret, frame = cap.read()
if not ret:
break
cv2_im = frame
cv2_im = imutils.resize(frame, width=640)
gray = cv2.cvtColor(cv2_im, cv2.COLOR_BGR2GRAY)
prev_yawn_status = yawnStatus
cv2_im_rgb = cv2.cvtColor(cv2_im, cv2.COLOR_BGR2RGB)
pil_im = Image.fromarray(cv2_im_rgb)
common.set_input(interpreter, pil_im)
interpreter.invoke()
objs = get_output(interpreter, score_threshold=args.threshold, top_k=args.top_k)
cv2_im = append_objs_to_img(cv2_im, objs, labels)
rects = detector(gray, 0)
for rect in rects:
shape = predictor(gray, rect)
shape = face_utils.shape_to_np(shape)
leftEye = shape[lStart:lEnd]
rightEye = shape[rStart:rEnd]
mouth = shape[mStart:mEnd]
leftEAR = eye_aspect_ratio(leftEye)
rightEAR = eye_aspect_ratio(rightEye)
mouEAR = mouth_aspect_ratio(mouth)
ear = (leftEAR + rightEAR) / 2.0
leftEyeHull = cv2.convexHull(leftEye)
rightEyeHull = cv2.convexHull(rightEye)
mouthHull = cv2.convexHull(mouth)
if ear < EYE_AR_THRESH:
COUNTER += 1
cv2.putText(cv2_im, "Eyes Closed ", (10, 30),cv2.FONT_HERSHEY_SIMPLEX, 0.7, (0, 0, 255), 2)
if COUNTER >= EYE_AR_CONSEC_FRAMES:
cv2.putText(cv2_im, "DROWSINESS ALERT!", (10, 50),
cv2.FONT_HERSHEY_SIMPLEX, 0.7, (0, 0, 255), 2)
if not ALARM_ON:
ALARM_ON = True
threadStatusQ.put(not ALARM_ON)
thread = Thread(target=soundAlert, args=(sound_path, threadStatusQ,))
thread.setDaemon(True)
thread.start()
else:
ALARM_ON=False
else:
COUNTER = 0
cv2.putText(cv2_im, "Eyes Open ", (10, 30),cv2.FONT_HERSHEY_SIMPLEX, 0.7, (0, 255, 0), 2)
if mouEAR > MOU_AR_THRESH:
cv2.putText(cv2_im, "Yawning ", (10, 70),cv2.FONT_HERSHEY_SIMPLEX, 0.7, (0, 0, 255),2)
yawnStatus = True
output_text = "Yawn Count: " + str(yawns + 1)
cv2.putText(cv2_im, output_text, (10,100),cv2.FONT_HERSHEY_SIMPLEX, 0.7,(255,0,0),2)
else:
yawnStatus = False
if prev_yawn_status == True and yawnStatus == False:
yawns+=1
cv2.imshow('frame', cv2_im)
if cv2.waitKey(1) & 0xFF == ord('q'):
break
cap.release()
cv2.destroyAllWindows()