def detect(self, image):
boxes = []
classes_id = []
confidences = []
scale = 0.00392
blob = cv2.dnn.blobFromImage(image,
scalefactor=scale,
size=(416, 416),
mean=(0, 0),
swapRB=True,
crop=False)
height, width = image.shape[:2]
# Lấy ảnh đầu vào cho model
self.model.setInput(blob)
# Thực hiện tính toán forward
outputs = self.model.forward(utils.get_output_layers(self.model))
# Xử lý các vùng có khả năng là biển số xe
for output in outputs:
for i in range(len(output)):
scores = output[i][5:]
class_id = np.argmax(scores)
confidence = float(scores[class_id])
# Chọn vùng có độ tin cậy rằng vùng đó khả năng lớn là biển số
if confidence > self.threshold:
# coordinate of bounding boxes
center_x = int(output[i][0] * width)
center_y = int(output[i][1] * height)
detected_width = int(output[i][2] * width)
detected_height = int(output[i][3] * height)
x_min = center_x - detected_width / 2
y_min = center_y - detected_height / 2
boxes.append(
[x_min, y_min, detected_width, detected_height])
classes_id.append(class_id)
confidences.append(confidence)
indices = cv2.dnn.NMSBoxes(boxes,
confidences,
score_threshold=self.threshold,
nms_threshold=0.4)
coordinates = []
# Lưu giá trị của các vùng chắc chắn là biển số
for i in indices:
index = i[0]
x_min, y_min, width, height = boxes[index]
x_min = round(x_min)
y_min = round(y_min)
coordinates.append((x_min, y_min, width, height))
return coordinates
def detect(self, image):
boxes = []
classes_id = []
confidences = []
scale = 0.00392
blob = cv2.dnn.blobFromImage(image,
scalefactor=scale,
size=(416, 416),
mean=(0, 0),
swapRB=True,
crop=False)
height, width = image.shape[:2]
# take image to model
self.model.setInput(blob)
# run forward
outputs = self.model.forward(utils.get_output_layers(self.model))
for output in outputs:
for i in range(len(output)):
scores = output[i][5:]
class_id = np.argmax(scores)
confidence = float(scores[class_id])
if confidence > self.threshold:
# coordinate of bounding boxes
center_x = int(output[i][0] * width)
center_y = int(output[i][1] * height)
detected_width = int(output[i][2] * width)
detected_height = int(output[i][3] * height)
x_min = center_x - detected_width / 2
y_min = center_y - detected_height / 2
boxes.append(
[x_min, y_min, detected_width, detected_height])
classes_id.append(class_id)
confidences.append(confidence)
indices = cv2.dnn.NMSBoxes(boxes,
confidences,
score_threshold=self.threshold,
nms_threshold=0.4)
coordinates = []
for i in indices:
index = i[0]
x_min, y_min, width, height = boxes[index]
x_min = round(x_min)
y_min = round(y_min)
coordinates.append((x_min, y_min, width, height))
return coordinates
def detect(self, img, bname, pixel_threshold=100000):
output_dir = "output/tmp"
boxes = []
classes_id = []
confidences = []
scale = 0.00392
blob = cv2.dnn.blobFromImage(img,
scalefactor=scale,
size=(416, 416),
mean=(0, 0, 0),
swapRB=True,
crop=False)
height, width = img.shape[:2]
# take image to model
self.model.setInput(blob)
# run forward
outputs = self.model.forward(utils.get_output_layers(self.model))
for out in outputs:
for detection in out:
scores = detection[5:]
class_id = np.argmax(scores)
confidence = float(scores[class_id])
if confidence > self.threshold:
# coordinate of bounding boxes
center_x = int(detection[0] * width)
center_y = int(detection[1] * height)
w = int(detection[2] * width)
h = int(detection[3] * height)
#cv2.circle(img,(center_x,center_y),10,(0,255,0),2)
#rectangle co-ordinaters
x = int(center_x - w / 2)
y = int(center_y - h / 2)
#cv2.rectangle(img,(x,y),(x+w,y+h),(0,255,0),2)
boxes.append([x, y, w, h])
classes_id.append(class_id)
confidences.append(confidence)
indices = cv2.dnn.NMSBoxes(boxes,
confidences,
score_threshold=self.threshold,
nms_threshold=0.3)
R = []
for i in range(len(boxes)):
if i in indices:
x, y, w, h = boxes[i]
label = str(self.labels[classes_id[i]])
R.append((label, confidences[i], (x, y, w, h)))
R = sorted(R, key=lambda x: -x[1])
R = [r for r in R if r[0] in ['car', 'bus']]
cars_img = []
Lcars = []
if len(R):
# Iorig = cv2.imread(img_path)
Iorig = img
WH = np.array(Iorig.shape[1::-1], dtype=float)
for i, r in enumerate(R):
# print("car size: w={}, h={}, area={}", r[2][2], r[2][3], r[2][2] * r[2][3])
_, _, ori_w, ori_h = r[2]
if ori_w * ori_h >= pixel_threshold:
cx, cy, w, h = (np.array(r[2]) / np.concatenate(
(WH, WH))).tolist()
tl = np.array([cx, cy])
br = np.array([cx + w, cy + h])
label = Label(0, tl, br)
Icar = crop_region(Iorig, label)
Lcars.append(label)
# cv2.imwrite('{}/{}_{}car.png'.format(output_dir, bname, i), Icar)
cars_img.append(Icar)
# lwrite('{}/{}_cars.txt'.format(output_dir, bname), Lcars)
print('\t\t{} cars found by using detect_vehicle'.format(
len(cars_img)))
return cars_img, Lcars