def stream_yolov3_personal(self):
net = ObjectDetection(conf_thresh=0.8, nms_thresh=0.4)
while True:
begin = process_time()
frame = Image.open(requests.get(self.url,
stream=True).raw).convert('RGB')
image, scale, padding = SingleImage(frame)[0]
image = torch.unsqueeze(image, 0)
detections = net.detect(image, scale, padding)
image_with_detections = net.draw_result(frame,
detections[0],
show=False)
opencvImage = cv2.cvtColor(np.array(image_with_detections),
cv2.COLOR_RGB2BGR)
end = process_time()
print(end - begin)
# show the frame to our screen
cv2.imshow("Frame", opencvImage)
key = cv2.waitKey(1) & 0xFF
# if the 'q' key is pressed, stop the loop
if key == ord("q"):
break
# close all windows
cv2.destroyAllWindows()
class PeopleDetection:
"""
The class recognizes people from the image and using face recognition detects their name
"""
def __init__(self):
self.obj_detect = ObjectDetection()
self.detect_face = FaceDetection()
self.reconize_face = FaceRecognition()
def detect(self, frame):
"""
Detect people
:param frame -- numpy image
:return -- details of each person(name, x, y, confidence, time_when_detected)
"""
details = []
detections = self.obj_detect.detect(frame)
# Get the objects detected by ObjectDetection
for i in range(detections.shape[0]):
confidence = detections[i, 2]
idx = int(detections[i, 1])
# If the object is a person and high confidence continue
if confidence > 0.25 and CLASSES[idx] == "person":
(h, w) = frame.shape[:2]
box = detections[i, 3:7] * np.array([w, h, w, h])
(startX, startY, endX, endY) = box.astype("int")
# In the person subframe detect faces(mostly should get 1 face)
sub_frame = frame[startY:endY, startX:endX, :]
face_locations = self.detect_face.detect(sub_frame)
area = float((endX - startX) * (endY - startY)) / (h * w)
# For each face get the person's name
for face_location in face_locations:
if face_location == ():
continue
(face_start_x, face_start_y, face_end_x,
face_end_y) = face_location
if face_end_x + startX > endX:
continue
# Draw the face
cv2.rectangle(
frame, (face_start_x + startX, face_start_y + startY),
(face_end_x + startX, face_end_y + startY),
(255, 0, 0), 2)
# Using Face recognition get the face and its score
name, score = self.reconize_face.face_recognize(
sub_frame, face_location)
# Important Conditions
face_h = face_end_y - face_start_y
face_w = face_end_x - face_end_y
y = startY - 15 if startY - 15 > 15 else startY + 15
start_x = np.floor(face_start_x - (face_w) / 4 + startX)
end_x = np.ceil(face_end_x + (face_w) / 4 + startX)
start_y = np.floor(face_end_y + 3 * (face_h) / 16 + startY)
end_y = np.ceil(face_end_x + 5 * (face_h) / 2 + startY)
t_start_x = int(start_x) if start_x > startX else startX
t_start_y = int(start_y) if start_y < endY else endY
t_end_x = int(end_x) if end_x < endX else endX
t_end_y = int(end_y) if end_y < endY else endY
cv2.rectangle(frame, (t_start_x, t_start_y),
(t_end_x, t_end_y), (255, 0, 0), 2)
# Area covered by the person of the screen
details.append({
'name':
name,
'x':
int((startX + endX) / 2),
'y':
int((startY + endY) / 2),
'score':
score,
'area':
area,
'time':
strftime("%Y-%m-%d %H:%M:%S", gmtime())
})
# Color detected person
sub_frame[:, :, 2] = sub_frame[:, :, 2] + 80
cv2.circle(frame, ((startX + endX) / 2, (startY + endY) / 2),
3, (0, 255, 0), -1)
# For no person detected, we require this for path planning
if len(details) == 0:
details.append({
'name': "No_person",
'x': int(480),
'y': int(270),
'score': 10.0,
'area': 0.0,
'time': strftime("%Y-%m-%d %H:%M:%S", gmtime())
})
cv2.imshow('frame', frame)
if cv2.waitKey(1) & 0xFF == ord('q'):
sys.exit(0)
return details
print('------------------------')
print('NUM:', count)
# 接收图片是否导入成功、帧图像
success, img = cap.read()
# 如果读入不到图像就退出
if not success:
break
center_points_current = [] # 储存当前帧的所有目标的中心点坐标
# 目标检测
# 将每一帧的图像传给目标检测方法
# 返回class_ids图像属于哪个分类；scores图像属于某个分类的概率；boxes目标检测的识别框
class_ids, scores, boxes = od.detect(img)
# 绘制检测框，boxes中包含每个目标检测框的左上坐标和每个框的宽、高
for box in boxes:
(x, y, w, h) = box
# 获取每一个框的中心点坐标，像素坐标是整数
cx, cy = int((x + x + w) / 2), int((y + y + h) / 2)
# 存放每一帧的所有框的中心点坐标
center_points_current.append((cx, cy))
# 绘制矩形框。传入帧图像，框的左上和右下坐标，框颜色，框的粗细
cv2.rectangle(img, (x, y), (x + w, y + h), (255, 0, 0), 2)
# 显示所有检测框的中心点，pt代表所有中心点坐标