def YOLO(video_path, filename):
# global metaMain, netMain, altNames
trash_set = set()
license_set = set()
start_time = time.time()
configPath = "./cfg/yolov4_trash.cfg"
weightPath = "./backup_trash/yolov4_trash_best.weights"
metaPath = "./cfg/trash.data"
if not os.path.exists(configPath):
raise ValueError("Invalid config path `" +
os.path.abspath(configPath) + "`")
if not os.path.exists(weightPath):
raise ValueError("Invalid weight path `" +
os.path.abspath(weightPath) + "`")
if not os.path.exists(metaPath):
raise ValueError("Invalid data file path `" +
os.path.abspath(metaPath) + "`")
m = Darknet(configPath)
m.print_network()
m.load_weights(weightPath)
print('Loading weights from %s... Done!' % (weightPath))
if use_cuda:
m.cuda()
num_classes = m.num_classes
namesfile = './cfg/trash.names'
class_names = load_class_names(namesfile)
fullpath = join(video_path, filename)
cap = CameraStream(fullpath).start()
fps = cap.get_fps() # OpenCV2 version 2 used "CV_CAP_PROP_FPS"
size = cap.get_size()
trash_max = 0
old_time = 0
# 記錄下來每個丟垃圾時間
trash_record_dict = {}
# 拆出時間
filename_start_time = filename[-10:-4]
# 分成 hour:minsec
filename_start_time = filename_start_time[:2] + ":" + filename_start_time[
2:]
# 分成 hour:min:sec
filename_start_time = filename_start_time[:5] + ":" + filename_start_time[
5:]
filename_tmp_x = time.strptime(
filename_start_time.split(',')[0], '%H:%M:%S')
filename_start_time_s = datetime.timedelta(
hours=filename_tmp_x.tm_hour,
minutes=filename_tmp_x.tm_min,
seconds=filename_tmp_x.tm_sec).total_seconds()
out = None
new_size = (416, 416)
while True:
prev_time = time.time()
ret, frame_read = cap.read()
old_frame = None
out_license = False
out_time = False
if ret:
old_frame = frame_read
frame_read = cv2.cvtColor(frame_read, cv2.COLOR_BGR2RGB)
else:
print('Video has ended or failed, try a different video format!')
break
frame_resized = cv2.resize(frame_read,
new_size,
interpolation=cv2.INTER_LINEAR)
boxes = do_detect(m, frame_resized, 0.4, 0.6, use_cuda)
trash_center = (0, 0)
license_center = (frame_resized.shape[0], frame_resized.shape[1])
frame_resized = cv2.resize(old_frame,
new_size,
interpolation=cv2.INTER_LINEAR)
width = frame_read.shape[1]
height = frame_read.shape[0]
for i in range(len(boxes[0])):
box = boxes[0][i]
# print(type(box))
# print(box)
xmin = int(box[0] * size[0])
ymin = int(box[1] * size[1])
xmax = int(box[2] * size[0])
ymax = int(box[3] * size[1])
cls_id = box[6]
# 畫圖
pt1 = (xmin, ymin)
pt2 = (xmax, ymax)
cv2.rectangle(frame_resized, pt1, pt2, (0, 255, 0), 1)
cv2.putText(
frame_resized,
class_names[cls_id] + " [" + str(round(box[5] * 100, 2)) + "]",
(pt1[0], pt1[1] - 5), cv2.FONT_HERSHEY_SIMPLEX, 0.5,
[0, 255, 0], 2)
# 確認是否為同個垃圾
center = (int((xmin + xmax) / 2), int((ymin + ymax) / 2))
temp = (center, pt1, pt2)
similar_trash = False
similar_license = False
if class_names[cls_id] == 'trash':
# temp = (center, pt1, pt2)
if temp not in trash_set:
# print('get new trash')
# out_time = True
for c in trash_set:
center_check = abs(c[0][0] -
center[0]) + abs(c[0][1] -
center[1])
if center_check <= 5:
similar_trash = True
break
# 交疊面積超過60%就當作同一個
else:
if (c[2][0] - xmin) + (c[2][1] - ymin) < (
xmax - xmin) + (ymax - ymin) + (
c[2][0] - c[1][0]) + (c[2][1] -
c[1][1]):
cross_xmin = max(xmin, c[1][0])
cross_ymin = max(ymin, c[1][1])
cross_xmax = min(xmax, c[2][0])
cross_ymax = min(ymax, c[2][1])
cross_area = abs((cross_xmax - cross_xmin) *
(cross_ymax - cross_ymin))
total_area = (xmax - xmin) * (ymax - ymin) + (
c[2][0] - c[1][0]) * (c[2][1] -
c[1][1]) - cross_area
if total_area != 0:
iou = float(cross_area / total_area)
if abs(iou) > 0.6:
similar_trash = True
break
if not similar_trash:
trash_set.add(temp)
out_time = True
else:
if temp not in license_set:
for c in license_set:
if abs(c[0][0] - center[0]) + abs(c[0][1] -
center[1]) <= 8:
similar_license = True
break
else:
if (c[2][0] - xmin) + (c[2][1] - ymin) < (
xmax - xmin) + (ymax - ymin) + (
c[2][0] - c[1][0]) + (c[2][1] -
c[1][1]):
cross_xmin = max(xmin, c[1][0])
cross_ymin = max(ymin, c[1][1])
cross_xmax = min(xmax, c[2][0])
cross_ymax = min(ymax, c[2][1])
cross_area = abs((cross_xmax - cross_xmin) *
(cross_ymax - cross_ymin))
total_area = (xmax - xmin) * (ymax - ymin) + (
c[2][0] - c[1][0]) * (c[2][1] -
c[1][1]) - cross_area
if total_area != 0:
iou = float(cross_area / total_area)
if abs(iou) > 0.8:
similar_license = True
break
if not similar_license:
license_set.add(temp)
out_license = True
result_img = plot_boxes_cv2(frame_resized,
boxes[0],
savename=None,
class_names=class_names)
frame_num = cap.get_frameNum()
duration = frame_num / fps
if out_time and duration > 5:
if duration > old_time + 10 or old_time == 0:
trash_max += 1
old_time = duration
if out_license:
Drop_trash_time_sec = str(
datetime.timedelta(seconds=(filename_start_time_s +
duration)))[:8]
if trash_max < 2:
trash_record_dict = {
filename[:-4] + "_Num_" + str(trash_max):
[Drop_trash_time_sec, str(1)]
}
else:
trash_record_dict[filename[:-4] + "_Num_" +
str(trash_max)] = [
Drop_trash_time_sec,
str(1)
]
resized = cv2.resize(result_img, (1280, 720),
interpolation=cv2.INTER_CUBIC)
cv2.imwrite(
'./output/image/' + filename[:-4] + '_' +
str(trash_max) + '_1.jpg', resized)
else:
Drop_trash_time_sec = str(
datetime.timedelta(seconds=(filename_start_time_s +
duration)))[:8]
if trash_max < 2:
trash_record_dict = {
filename[:-4] + "_Num_" + str(trash_max):
[Drop_trash_time_sec, str(0)]
}
else:
trash_record_dict[filename[:-4] + "_Num_" +
str(trash_max)] = [
Drop_trash_time_sec,
str(0)
]
resized = cv2.resize(result_img, (1280, 720),
interpolation=cv2.INTER_CUBIC)
cv2.imwrite(
'./output/image/' + filename[:-4] + '_' +
str(trash_max) + '_0.jpg', resized)
# print('trash_record_dict', trash_record_dict)
out_time, out_license = False, False
# cv2.putText(image, "Catch number " + str(trash_max), (10,40), cv2.FONT_HERSHEY_COMPLEX_SMALL, 1, [0, 0, 0], 2)
cap.stop()
if len(trash_record_dict) > 0:
Drop_trash_df = pd.DataFrame(list(trash_record_dict.items()),
columns=['Filename', 'Record'])
Drop_trash_df[['RecordTime', 'Licence'
]] = pd.DataFrame(Drop_trash_df.Record.tolist(),
index=Drop_trash_df.index)
Drop_trash_df.drop(columns=["Record"], inplace=True)
print(Drop_trash_df)
Drop_trash_df.to_csv("./output/csv_file/" + filename[:-4] + ".csv",
index=False)
# print('#############start Record Video#################')
# # filename ='20200927_234800.avi'
# Trash_df = pd.read_csv("./trash_data/record/csv_file/"+filename[:-4]+".csv")
# filename_start_time = filename[-10:-4]
# filename_start_time = filename_start_time[:2] + ":" + filename_start_time[2:]
# filename_start_time = filename_start_time[:5] + ":" + filename_start_time[5:]
# filename_tmp_x = time.strptime(filename_start_time.split(',')[0][:-1],'%H:%M:%S')
# filename_start_time_s = datetime.timedelta(hours=filename_tmp_x.tm_hour,minutes=filename_tmp_x.tm_min,seconds=filename_tmp_x.tm_sec).total_seconds()
# print(Trash_df)
# for i in range(len(Trash_df)):
# fullpath = join(video_path, filename)
# video = cv2.VideoCapture(fullpath)
# frame_num = 0
# frame_width = int(video.get(3))
# frame_height = int(video.get(4))
# size = (frame_width, frame_height)
# result = cv2.VideoWriter('./trash_data/record/video/' + filename[:-4] + '_' + listToString(Trash_df['RecordTime'][i].split(':')) + '_1.avi',
# cv2.VideoWriter_fourcc(*'MJPG'),
# 10, size)
# x = time.strptime(Trash_df['RecordTime'][i].split(',')[0][:-1],'%H:%M:%S')
# y = datetime.timedelta(hours=x.tm_hour,minutes=x.tm_min,seconds=x.tm_sec).total_seconds()
# y_add_10_sec = y + 10
# y_minus_10_sec = y - 10
# y_add_10_sec = y_add_10_sec - filename_start_time_s
# y_minus_10_sec = y_minus_10_sec - filename_start_time_s
# str_add_10_sec = str(datetime.timedelta(seconds=y_add_10_sec))
# str_minus_10_sec = str(datetime.timedelta(seconds=y_minus_10_sec))
# print('Cuted Video Enging time ', str_add_10_sec, 'Cuted Video Begin time ', str_minus_10_sec)
# while(True):
# ret, frame = video.read()
# frame_num += 1
# if ret == True:
# duration = frame_num/fps
# duration_time = str(datetime.timedelta(seconds = (duration)))[:7]
# if duration_time >= str_minus_10_sec and duration_time <= str_add_10_sec and frame_num%2 == 0:
# # print('Save....', duration_time)
# result.write(frame)
# else :
# print('End ret')
# break
# video.release()
# result.release()
# # When everything done, release
# # the video capture and video
# # write objects
# print("The video was successfully saved")
else:
print('Catch Nothing')
print('How much time we spend for this video? ' +
str(time.time() - start_time))
print(
"*****************************End*************************************"
)
class TestView(object):
def __init__(self):
self.client = mqtt.Client()
self.client.connect('xxx.xxx.xxx.xxx', 8084)
# predict 用的資訊
# This is needed since the notebook is stored in the object_detection folder.
sys.path.append("..")
self.cap = None
# Name of the directory containing the object detection module we're using
MODEL_NAME = 'inference_graph'
fontPath = "./TaipeiSansTCBeta-Bold.ttf"
self.font = ImageFont.truetype(fontPath, 32)
# Grab path to current working directory
CWD_PATH = os.getcwd()
# Path to frozen detection graph .pb file, which contains the model that is used
# for object detection.
PATH_TO_CKPT = os.path.join(CWD_PATH, MODEL_NAME,
'frozen_inference_graph.pb')
# Path to label map file
PATH_TO_LABELS = os.path.join(CWD_PATH, 'training', 'labelmap.pbtxt')
# Number of classes the object detector can identify
NUM_CLASSES = 20
## Load the label map.
# Label maps map indices to category names, so that when our convolution
# network predicts `5`, we know that this corresponds to `king`.
# Here we use internal utility functions, but anything that returns a
# dictionary mapping integers to appropriate string labels would be fine
label_map = label_map_util.load_labelmap(PATH_TO_LABELS)
categories = label_map_util.convert_label_map_to_categories(
label_map, max_num_classes=NUM_CLASSES, use_display_name=True)
self.category_index = label_map_util.create_category_index(categories)
for key in self.category_index:
print("key: %s , value: %s" % (key, self.category_index[key]))
# Load the Tensorflow model into memory.
detection_graph = tf.Graph()
with detection_graph.as_default():
od_graph_def = tf.GraphDef()
with tf.gfile.GFile(PATH_TO_CKPT, 'rb') as fid:
serialized_graph = fid.read()
od_graph_def.ParseFromString(serialized_graph)
tf.import_graph_def(od_graph_def, name='')
self.sess = tf.Session(graph=detection_graph)
# Define input and output tensors (i.e. data) for the object detection classifier
# Input tensor is the image
self.image_tensor = detection_graph.get_tensor_by_name(
'image_tensor:0')
# Output tensors are the detection boxes, scores, and classes
# Each box represents a part of the image where a particular object was detected
self.detection_boxes = detection_graph.get_tensor_by_name(
'detection_boxes:0')
# Each score represents level of confidence for each of the objects.
# The score is shown on the result image, together with the class label.
self.detection_scores = detection_graph.get_tensor_by_name(
'detection_scores:0')
self.detection_classes = detection_graph.get_tensor_by_name(
'detection_classes:0')
# Number of objects detected
self.num_detections = detection_graph.get_tensor_by_name(
'num_detections:0')
class MyEncoder(json_dict.JSONEncoder):
def default(self, obj):
if isinstance(obj, np.integer):
return int(obj)
elif isinstance(obj, np.floating):
return float(obj)
elif isinstance(obj, np.ndarray):
return obj.tolist()
else:
return super(MyEncoder, self).default(obj)
def mqtt_pub(self, payload):
topic = 'demo/fish'
json_out = {}
json_out['GwID'] = "MiGW-B1-1"
json_out['TimeStamp'] = round(time.time() * 1000)
json_out['DeviceID'] = "B1F-IPCam"
json_out['Payload'] = payload
json_str = json_dict.dumps(json_out, cls=self.MyEncoder)
self.client.publish(topic, json_str, 0, False)
def predict(self, frame):
if frame is None:
# print("The read queue is None, plz do not predict!")
return None
frame_expanded = np.expand_dims(frame, axis=0)
#print(frame_expanded.shape)
# Perform the actual detection by running the model with the image as input
(boxes, scores, classes,
num) = self.sess.run([
self.detection_boxes, self.detection_scores,
self.detection_classes, self.num_detections
],
feed_dict={self.image_tensor: frame_expanded})
vis_util.visualize_boxes_and_labels_on_image_array(
frame,
np.squeeze(boxes),
np.squeeze(classes).astype(np.int32),
np.squeeze(scores),
self.category_index,
use_normalized_coordinates=True,
line_thickness=3,
skip_labels=True,
skip_scores=True,
min_score_thresh=0.60)
payload = []
for i in range(len(scores[0])):
if scores[0][i] > 0.6:
# payload design
temp_list = {}
# temp_list.setdefault('id', str(i))
temp_list.setdefault('score', str(scores[0][i]))
temp_list.setdefault('class', str(classes[0][i]))
temp_list.setdefault(
'name', self.category_index[classes[0][i]]['name'])
temp_list.setdefault('ymin', boxes[0][i][0])
# print("這是還沒json的座標 ", boxes[j][i][0], " 跟 ", boxes[j][i][1])
temp_list.setdefault('xmin', boxes[0][i][1])
temp_list.setdefault('ymax', boxes[0][i][2])
temp_list.setdefault('xmax', boxes[0][i][3])
payload.append(temp_list)
self.mqtt_pub(payload)
convert = cv2.imencode('.jpg', frame)[1].tobytes()
return convert
def gen_frame(self):
"""
Video stream generator
"""
self.cap = CameraStream().start()
while self.cap:
frame = self.cap.read()
convert = self.predict(frame)
if convert is None:
continue
yield (b'--frame\r\n'
b'Content-Type: image/jpeg\r\n\r\n' + convert + b'\r\n'
) # concate frame one by one and show result
def __del__(self):
try:
self.cap.stop()
self.cap.stream.release()
except:
print('probably there\'s no cap yet :(')
cv2.destroyAllWindows()
self.client.disconnect()
import cv2
from camera import CameraStream
if __name__ == "__main__":
cam = CameraStream().start()
while True:
frame = cam.read()
cv2.imshow('webcam', frame)
if cv2.waitKey(1) == 27:
break
cam.stop()
cv2.destroyAllWindows()