def runDetectionsOnCam(url, camName):
global config
newFrame = None
detector = detect.detector(camName, config.get('ssd', 'protocol'),
config.get('ssd', 'model'))
##detector = detect_motion.detect_motion()
cam = webcam.threadCamReader(url)
cam.start()
sender = imagezmq.ImageSender(connect_to='tcp://*:555' + camName[-1:],
block=False)
readFrameID = None
while True:
time.sleep(1 / 10000)
frame, frameID = cam.read()
#Just skip if we read the same frame or done have a frame at all
if frame is None or frameID is None or readFrameID == frameID:
#print(camName + ' ' + str(frameID))
continue
readFrameID = frameID
newFrame = detector.detect(frame, config.get(camName, 'fps'))
if newFrame is not None:
sender.send_image(camName, newFrame)
def realtime():
#initialize preview
cv2.namedWindow("preview")
dtr = detector()
vc = cv2.VideoCapture(0)
model = load_model()
max_X = 20 # max movement of x possible
max_Y = 20 # max movement of y possible
min_jump_X = 600 # not used now
min_jump_Y = 400 # not used now
prev_X, prev_Y = locate_cursor() # initial position
if vc.isOpened(): #get the first frame
rval, frame = vc.read()
else:
rval = False
while rval:
frame = cv2.flip(frame, 1)
eyes = dtr.detect(frame)
for (x, y, w, h) in eyes:
img = cv2.cvtColor(frame[y:y + h, x:x + w], cv2.COLOR_RGB2GRAY)
img = cv2.resize(img, (50, 50), interpolation=cv2.INTER_AREA)
frame = cv2.rectangle(frame, (x, y), (x + w, y + h), (255, 0, 0),
2)
#cv2.circle(frame,(int(x + w/2),int(y+h/2)),int(h*0.7),(255,0,0),2)
img = img.reshape((1, ) + img.shape + (1, )) / 255.
y_pred = model.predict(img) #predict the position
x = int(y_pred[0][0])
y = int(y_pred[0][1])
if x - prev_X > max_X:
x = prev_X + max_X
elif x - prev_X < -max_X:
x = prev_X - max_X
if y - prev_Y > max_Y:
y = prev_Y + max_Y
elif y - prev_Y < -max_Y:
y = prev_Y - max_Y
move(x, y)
prev_X = x
prev_Y = y
break
frame = cv2.resize(frame, (200, 160), interpolation=cv2.INTER_AREA)
cv2.imshow('preview', frame)
rval, frame = vc.read()
key = cv2.waitKey(20)
if key == 27: # exit on ESC
break
cv2.destroyWindow("preview")
vc = None
def generate_data():
cursor = []
i = 20600
cv2.namedWindow("preview")
dtr = detector()
vc = cv2.VideoCapture(0)
if vc.isOpened(): #get the first frame
rval, frame = vc.read()
else:
rval = False
while rval:
frame = cv2.flip(frame, 1)
eyes = dtr.detect(frame)
# if len(eyes) > 0:
# x,y,w,h = eyes[0]
# frame = cv2.rectangle(frame,(x,y),(x+w,y+h),(255,0,0),2)
# eye = frame[y:y+h, x:x+w]
# eye = cv2.cvtColor( eye, cv2.COLOR_RGB2GRAY )
# eye = cv2.resize(eye, (50, 50))
# cv2.imwrite('.\\Data\\eye_images\\eye_{}.jpg'.format(i), eye)
# cursor.append(get_cursor_position())
# i += 1
for (x, y, w, h) in eyes:
frame = cv2.rectangle(frame, (x, y), (x + w, y + h), (255, 0, 0),
2)
eye = frame[y:y + h, x:x + w]
eye = cv2.cvtColor(eye, cv2.COLOR_RGB2GRAY)
eye = cv2.resize(eye, (50, 50))
cv2.imwrite('.\\Data\\eye_images\\eye_{}.jpg'.format(i), eye)
cursor.append(get_cursor_position())
i += 1
print(i)
break
cv2.imshow('preview', frame)
rval, frame = vc.read()
key = cv2.waitKey(20)
if key == 27: # exit on ESC
break
cv2.destroyWindow("preview")
vc = None
return cursor
def find_obj(core):
'''
Rotates right until an object is detected by obj detector
'''
obj_detected = False
while not obj_detected:
send_act(core, 'TURN_RIGHT')
time.sleep(.4)
send_act(core, 'STOP')
time.sleep(.01) #stop briefly to stablize camera image
#Obj detected?
img = take_image()
if img is not None:
#find if object is detected
obj_detected, obj_img = detector(img)
return obj_img
def __init__(self):
# Deploy darknet53 model on cooresponding device
yolov3 = darknet("cfg/yolov3-1.cfg", 1)
yolov3.load_weight("yolov3-1.weights")
# Deploy stacked hourglass model
stackedhourglass = StackedHourglass(16)
stackedhourglass.load_state_dict(torch.load("stacked_hourglass.pkl"))
cuda = torch.cuda.is_available()
if cuda:
yolov3.cuda()
stackedhourglass.cuda()
yolov3.eval()
self.detector = detector(yolov3)
self.estimator = Estimator(stackedhourglass)
def collect_images(core, num_images):
'''
Rotates and saves images.
'''
for i in range(num_images):
while True: #primary loop
send_act(core, 'TURN_RIGHT')
time.sleep(.4)
send_act(core, 'STOP')
time.sleep(.01)
img = take_image()
if img is not None:
obj_detected, obj_img = detector(img)
save_image('images/image'+str(i)+'.jpg', img)
if obj_detected:
save_image('images/detected'+str(i)+'.jpg', obj_img)
i+=1
time.sleep(.01)
def __init__(self):
# Deploy darknet53 model on cooresponding device
yolov3 = darknet("cfg/yolov3.cfg", 80)
yolov3.load_weight("yolov3.weights")
yolov3.eval()
# Deploy stacked hourglass model
stackedhourglass = demo.__dict__['hg'](num_stacks=2,
num_blocks=1,
num_classes=16)
stackedhourglass = torch.nn.DataParallel(stackedhourglass)
stackedhourglass.eval()
checkpoint = torch.load('demo/hg_s2_b1/model_best.pth.tar')
stackedhourglass.load_state_dict(checkpoint['state_dict'])
cuda = torch.cuda.is_available()
if cuda:
yolov3.cuda()
stackedhourglass.cuda()
self.detector = detector(yolov3)
self.estimator = stackedhourglass
)
st.write(
'Welcome to the 2MCAM web application that can accurately identify human beings that violate social distancing norms in public spaces.'
)
elif option == 'Social distancing detector':
st.title('Social Distancing Detection')
st.write(
'The practice of social distancing signifies maintaining a distance of 6 feet or more when in public places or simply staying at home and away from others as much as possible to help prevent spread of COVID-19.'
)
st.write(
'A green/red bounding box is drawn over the individuals in the frame to approve or flag the distance between them. '
)
if st.button("Start"):
detector()
elif option == 'Learn more!':
st.title('Why 2MCAM?')
st.image('data/img1.jpg')
st.write(
'2MCAM is a user-friendly ML web application and is designed to detect possible violations of social distancing norms.'
)
st.write(
'Violation of physical distancing norms (1.5 - 2 meters) is looked down upon in the era of the covid19 pandemic. This is why we have a solution for you: 2MCAM, the web application that can instantaneously detect and flag such violations.'
)
st.image('data/2feet.jpeg')
st.write(
'Our application has a light-weight frontend and heavily tested backend waiting to be used. Discover the working of our application by navigating to the social distancing detector section.'
)
def run(self):
idx_frame = 0
# Load yolov3_tiny_se detect
weights = 'best4.pt'
cfg = 'yolov3-tiny-1cls-se.cfg'
img_size = 416
device = torch_utils.select_device(device='0')
# print(device)
# Initialize model
model = Darknet(cfg, img_size)
# Load weights
attempt_download(weights)
if weights.endswith('.pt'): # pytorch weights format
model.load_state_dict(
torch.load(weights, map_location=device)['model'])
else:
load_darknet_weights(model, weights)
# 预载模型
model.to(device).eval()
fps_list = []
while self.vdo.grab():
idx_frame += 1
if idx_frame % self.args.frame_interval:
continue
_, ori_im = self.vdo.read() # get frame
if ori_im is not None:
im = cv2.cvtColor(ori_im, cv2.COLOR_BGR2RGB) # init frame
#print(im.shape)
start = time.time()
# 获取 detection后的结果
bbox_xywh, cls_conf, cls_ids, bbox_xyxy1 = detector(
im, device, model)
#print(bbox_xywh, cls_conf, cls_ids)
stop = time.time()
if bbox_xywh is not None:
# select car class
mask = cls_ids == 0
bbox_xywh = bbox_xywh[mask]
bbox_xywh[:, 3:] *= 1.2 # 放大 bbox，避免 bbox过小
cls_conf = cls_conf[mask]
# 获取目标 tracking
outputs, sum_id, track_list = self.deepsort.update(
bbox_xywh, cls_conf, im)
# yolov3 和 DeepSort的可视化
if len(outputs) > 0:
bbox_xyxy = outputs[:, :4]
identities = outputs[:, -1]
# print(bbox_xyxy,identities)
ori_im = draw_boxes(ori_im, bbox_xyxy, track_list,
identities) # DeepSort
ori_im = draw_boxes111(ori_im, bbox_xyxy1,
cls_conf) # YOLOv3
# ori_im = draw_track(ori_im, bbox_xyxy, track_list) # track
if bbox_xywh is None:
sum_car = 'Traffic flow(frame): ' + '0'
ori_im = cv2.putText(ori_im, sum_car, (10, 50),
cv2.FONT_HERSHEY_PLAIN, 2.5,
[255, 255, 255], 2)
ori_im = draw_id(ori_im, sum_id) # sum_id
end = time.time()
fps = 1 / (end - start + 0.001)
fps_list.append(fps)
# print("yolov3_tiny-time: {:.03f}s, fps: {:.03f}".format(stop - start, 1 / (stop - start)))
print("total-time: {:.03f}s, fps: {:.03f}".format(
end - start, fps))
if self.args.display and ori_im is not None:
cv2.imshow("test", ori_im)
cv2.waitKey(1)
if self.args.save_path:
self.writer.write(ori_im)
avg_fps = np.mean(fps_list)
print("avg_fps: {:.03f}".format(avg_fps))
device = torch_utils.select_device(device='0')
print(device)
# Initialize model
model = Darknet(cfg, img_size)
# Load weights
attempt_download(weights)
if weights.endswith('.pt'): # pytorch format
model.load_state_dict(torch.load(weights, map_location=device)['model'])
else: # darknet format
load_darknet_weights(model, weights)
# Eval mode
model.to(device).eval()
while True:
ret, frame = cap.read()
if ret:
print(frame.shape)
start = time.time()
bbox_xywh, cls_conf, cls_ids = detector(frame, device, model)
end = time.time()
print(bbox_xywh, cls_conf, cls_ids)
print(end - start, 1 / (end - start + 0.00000001))
else:
break
cap.release()
cv2.destroyAllWindows()
def run(self):
idx_frame = 0
# Load yolov3_tiny_se detect
weights = 'best4.pt'
cfg = 'yolov3-tiny-1cls-se.cfg'
img_size = 416
device = torch_utils.select_device(device='0')
#print(device)
# Initialize model
model = Darknet(cfg, img_size)
# Load weights
attempt_download(weights)
if weights.endswith('.pt'): # pytorch format
model.load_state_dict(
torch.load(weights, map_location=device)['model'])
else: # darknet format
load_darknet_weights(model, weights)
# Eval mode
model.to(device).eval()
fps_list = []
while self.vdo.grab():
idx_frame += 1
if idx_frame % self.args.frame_interval:
continue
_, ori_im = self.vdo.retrieve()
im = cv2.cvtColor(ori_im, cv2.COLOR_BGR2RGB)
#print(im.shape)
start = time.time()
# do detection
bbox_xywh, cls_conf, cls_ids, bbox_xyxy1 = detector(
im, device, model)
#print(bbox_xywh, cls_conf, cls_ids)
stop = time.time()
if bbox_xywh is not None:
# select car class
mask = cls_ids == 0
bbox_xywh = bbox_xywh[mask]
#bbox_xywh[:,3:] *= 1.2 # bbox dilation just in case bbox too small
cls_conf = cls_conf[mask]
# do tracking
outputs = self.deepsort.update(bbox_xywh, cls_conf, im)
# draw boxes for visualization
if len(outputs) > 0:
bbox_xyxy = outputs[:, :4]
identities = outputs[:, -1]
#print(bbox_xyxy,identities)
ori_im = draw_boxes(ori_im, bbox_xyxy, identities)
ori_im = draw_boxes111(ori_im, bbox_xyxy1, cls_conf)
end = time.time()
fps = 1 / (end - start + 0.001)
fps_list.append(fps)
#print("yolov3_tiny-time: {:.03f}s, fps: {:.03f}".format(stop - start, 1 / (stop - start)))
print("total-time: {:.03f}s, fps: {:.03f}".format(
end - start, fps))
if self.args.display:
cv2.imshow("test", ori_im)
cv2.waitKey(1)
if self.args.save_path:
self.writer.write(ori_im)
avg_fps = np.mean(fps_list)
print("avg_fps: {:.03f}".format(avg_fps))
def recognition():
detector()