def __init__(self):
self.yolo_anchors_path = yaml_cfg.get('yolo_anchors_path')
self.yolo_model_path = yaml_cfg.get('yolo_model_path')
self.yolo_classes_path = yaml_cfg.get('yolo_classes_path')
self.usr_model_type = yaml_cfg.get('usr_model_type')
if self.usr_model_type == 'yolo':
logger.info("use yolo model in human detecting")
self.yolo = YOLO(self.yolo_anchors_path, self.yolo_model_path,
self.yolo_classes_path)
else:
logger.info("use pose model in human detecting")
w, h = networks.model_wh(yaml_cfg.get('pose_image_size'))
self.poseEstimator = TfPoseEstimator(
networks.get_graph_path('cmu'), target_size=(w, h))
def mod_ins_stuff(self, empe_inf):
mnt_tp_cd = empe_inf[constants.cons_request_mnt_tp_cd]
tmp_emb = None
if mnt_tp_cd == constants.cons_mnt_tp_cd_delete:
self.__mod_ins_stuff(mnt_tp_cd,
empe_inf[constants.cons_request_ccbins_id],
empe_inf[constants.cons_request_empe_id],
empe_inf[constants.cons_request_usr_nm], None)
else:
# 将图象由OpenCv的BGR转成RGB格式，wuyunzhen.zh, 20190309
img = cv.cvtColor(empe_inf[constants.cons_request_img],
cv.COLOR_BGR2RGB)
tmp_embs = self.get_embedding(img)
if len(tmp_embs) > 0:
tmp_emb = tmp_embs[0]
self.__mod_ins_stuff(
mnt_tp_cd, empe_inf[constants.cons_request_ccbins_id],
empe_inf[constants.cons_request_empe_id],
empe_inf[constants.cons_request_usr_nm], tmp_emb)
# 将图片保存到磁盘文件
img_path = yaml_cfg.get('ccbins_stuff_pcs_imgs') + "/" + empe_inf[constants.cons_request_ccbins_id] + '_' + \
empe_inf[constants.cons_request_empe_id] + '_' + empe_inf[constants.cons_request_usr_nm] + ".jpg"
cv.imwrite(img_path, empe_inf[constants.cons_request_img])
else:
raise AiException(HttpStatus.SERVER_ERROR,
AiErrorCode.YCEA4021011)
return tmp_emb
def humanDetect(self, req_data, cv_image_list, path_flag):
results = []
write_dbs = []
cur_date = time.strftime('%Y%m%d', time.localtime(time.time()))
ccbins_id = req_data.get(constants.cons_request_ccbins_id)
sys_evt_trace_id = req_data.get(constants.cons_request_trace_id)
folder_path = yaml_cfg.get('illegal_image_path') + "/" + cur_date +'/'+ ccbins_id
save_all_file = yaml_cfg.get('save_all_file')
if not os.path.exists(folder_path):
os.makedirs(folder_path)
folder_path = os.path.abspath(folder_path)
for i in range(0, len(cv_image_list)):
face_count = 0
#人数、非法人数为-1，表示该项无须校验
human_count = -1
un_num = -1
valid_stff_list = []
for type in req_data.get(constants.cons_request_model_type):
if type == constants.cons_mdtp_human_detect:
# 人数检测
human_count = self.humanCountSrv.humanCount(cv_image_list[i])
elif type == constants.cons_mdtp_human_recg:
# 非法入侵
face_count, un_num, valid_stff_list = self.humanRecognizeSrv.get_invalid_face(cv_image_list[i], ccbins_id)
results.append([human_count, face_count, un_num, valid_stff_list])
detect_alarm_flag, recognize_alarm_flag = self.__computeAlarm(human_count, un_num)
img_path = (req_data.get(constants.cons_request_paths)[i] if path_flag else (
folder_path + "/" + sys_evt_trace_id + "_" + str(i) + ".jpg"))
if not path_flag:
if detect_alarm_flag or recognize_alarm_flag or save_all_file:
# 若不是通过交易带附件传图片且初步认为有告警产生，则保存图片
cv.imwrite(img_path, cv_image_list[i])
# 将结果写数据库
write_dbs.append([human_count, un_num, valid_stff_list, img_path, detect_alarm_flag, recognize_alarm_flag])
oprt_rcd_dao.add_records(write_dbs, req_data)
return results
def __init__(self):
# 扣人脸模型初始化
self.minsize = yaml_cfg.get('facenet_min_size') # minimum size of face
# three steps's threshold
self.threshold = list(
map(float,
yaml_cfg.get('facenet_threshold').split(',')))
self.factor = float(yaml_cfg.get('facenet_factor')) # scale factor
self.margin = 44
self.image_size = int(yaml_cfg.get('facenet_image_size'))
self.graph_find = tf.Graph()
self.facenet_confidence_level = float(
yaml_cfg.get('facenet_confidence_level'))
self.facenet_ingore_small = yaml_cfg.get('facenet_ingore_small')
self.head_pitch_angle = 35 # 人员点头最大角度
self.head_yaw_angle = 35 # 人员侧脸最大角度
self.head_pitch_yaw_angle = 60 # 人员点后/侧脸角度和最大值
self.face_color_percent = 0.6 # 符合人脸颜色的像素占整个人脸区域的比重最小值
with self.graph_find.as_default():
# gpu_options = tf.GPUOptions(per_process_gpu_memory_fraction=0.3)
# self.sess_find = tf.Session(config=tf.ConfigProto(gpu_options=gpu_options, log_device_placement=False))
self.sess_find = tf.Session()
with self.sess_find.as_default():
self.pnet, self.rnet, self.onet = detect_face.create_mtcnn(
self.sess_find, None)
# 识别人脸模型初始化
self.graph_recognize = tf.Graph()
with self.graph_recognize.as_default():
# gpu_options = tf.GPUOptions(per_process_gpu_memory_fraction=0.3)
# self.sess_recognize = tf.Session(config=tf.ConfigProto(gpu_options=gpu_options, log_device_placement=False))
self.sess_recognize = tf.Session()
with self.sess_recognize.as_default():
# Load the model
# facenet.load_model("20170512-110547")
saver = tf.train.import_meta_graph(
"apps/humanDetectApp/models/facenet/20170512-110547/model-20170512-110547.meta"
)
saver.restore(
tf.get_default_session(),
"apps/humanDetectApp/models/facenet/20170512-110547/model-20170512-110547.ckpt-250000"
)
# saver = tf.train.import_meta_graph(
# "apps/humanDetectApp/models/facenet/20180402-114759/model-20180402-114759.meta")
# saver.restore(tf.get_default_session(),
# "apps/humanDetectApp/models/facenet/20180402-114759/model-20180402-114759.ckpt-275")
# Get input and output tensors
self.images_placeholder = tf.get_default_graph(
).get_tensor_by_name("input:0")
self.embeddings = tf.get_default_graph().get_tensor_by_name(
"embeddings:0")
self.phase_train_placeholder = tf.get_default_graph(
).get_tensor_by_name("phase_train:0")
self.legal_ins_stuff = {}
self.reg_ins_stuff()
def __reg_from_file(self):
stuff_image_dir = yaml_cfg.get('ccbins_stuff_imgs')
list = os.listdir(stuff_image_dir) # 列出文件夹下所有的目录与文件
success = 0
for i in range(0, len(list)):
path = os.path.join(stuff_image_dir, list[i])
if os.path.isfile(path):
img = cv.imread(path)
# cv.imwrite("src_file.jpg", img)
img = cv.cvtColor(img, cv.COLOR_BGR2RGB)
# cv.imwrite("src_file_rgb.jpg", img)
embs = self.get_embedding(img)
if len(embs) > 0:
success += 1
ccbinsId, empeId, usrNm = self.__getStuffInfoByPath(path)
tmp_ccbins = []
for (key, value) in self.legal_ins_stuff.items():
if key == ccbinsId:
tmp_ccbins = value
if tmp_ccbins:
tmp_ccbins.append({
constants.cons_request_empe_id: empeId,
constants.cons_request_usr_nm: usrNm,
constants.cons_emb: embs[0]
})
else:
self.legal_ins_stuff.update({
ccbinsId: [{
constants.cons_request_empe_id: empeId,
constants.cons_request_usr_nm: usrNm,
constants.cons_emb: embs[0]
}]
})
else:
logger.error("图片中没有人脸, 路径:" + path)
# print(self.legal_ins_stuff)
logger.info("从配置文件注册人脸成功，人脸数: " + str(len(list)) + "，成功数量:" +
str(success))
# coding=utf-8
import apps.humanDetectApp.models.pose_estimation.src.networks as networks
from apps.humanDetectApp.models.pose_estimation.src.estimator import TfPoseEstimator
from networks.yolo.yolo3_predict import YOLO
from PIL import Image
from common import addlog, ai_print
import tensorflow as tf
import logging.config
from keras import backend as K
from apps.humanDetectApp.human_detect_config import yaml_cfg
logger = logging.getLogger()
config = tf.ConfigProto()
config.gpu_options.per_process_gpu_memory_fraction = float(
yaml_cfg.get('per_process_gpu_memory_fraction'))
session = tf.Session(config=config)
K.set_session(session)
class HumanCountSrv(object):
def __init__(self):
self.yolo_anchors_path = yaml_cfg.get('yolo_anchors_path')
self.yolo_model_path = yaml_cfg.get('yolo_model_path')
self.yolo_classes_path = yaml_cfg.get('yolo_classes_path')
self.usr_model_type = yaml_cfg.get('usr_model_type')
if self.usr_model_type == 'yolo':
logger.info("use yolo model in human detecting")
self.yolo = YOLO(self.yolo_anchors_path, self.yolo_model_path,
self.yolo_classes_path)
else:
logger.info("use pose model in human detecting")
def __computeAlarm(self, human_count, un_num):
legal_stuff_num = yaml_cfg.get('legal_stuff_num')
return (human_count != -1 and human_count != 0 and human_count < legal_stuff_num), un_num > 0
def post(self):
orgName = get_argument("orgName", required=True, help='所属机构不能为空')
deviceName = get_argument("deviceName", required=True, help='所属设备不能为空')
alarmType = get_argument("alarmType", required=True, help='报警类型不能为空')
alarmLevel = get_argument("alarmLevel", required=True, help='报警等级不能为空')
alarmTime = get_argument("alarmTime", required=True, help='报警时间不能为空')
remark = get_argument("remark", required=False)
# 转回img
img = pickle.loads(eval(remark))
# json_data = request.form['json_data']
# dict_data = json.loads(json_data)
# orgName =dict_data.get('orgName',-1)
# deviceName =dict_data.get('deviceName',-1)
# alarmType =dict_data.get('alarmType')
# alarmLevel =dict_data.get('alarmLevel')
# alarmTime =dict_data.get('alarmTime')
# f = request.files['myfile']
# x=f.read()
# print('>>>',x)
# with open('D:/image/b123.jpg', 'ab') as f:
# f.write(x)
# for i in f:
# print('@@',type(i))
# with open('D:/image/a.jpg', 'ab') as f:
# f.write(i)
# 告警类别 00 对应海康 人数超限：'131643', '单人异常', '131644', '多人异常'
# 告警类别 01 对应海康 非法入侵：'131585', '跨域警戒线','131588', '区域入侵', '131586', '人员进入'
if alarmType == '131643':
alarmType = '00'
vlt_err_alrm_inf = '单人异常'
elif alarmType == '131644':
alarmType = '00'
vlt_err_alrm_inf = '多人异常'
elif alarmType == '131585':
alarmType = '01'
vlt_err_alrm_inf = '跨域警戒线'
elif alarmType == '131588':
alarmType = '01'
vlt_err_alrm_inf = '区域入侵'
elif alarmType == '131586':
alarmType = '01'
vlt_err_alrm_inf = '人员进入'
elif alarmType == '150002':
alarmType = '01'
vlt_err_alrm_inf = '非法人员入侵'
else:
raise AiException(HttpStatus.SERVER_ERROR, AiErrorCode.YCEA4021012)
alarm_info_hk_po = AlarmInfoHk()
alarm_info_hk_po.id = gen_uuid()
alarm_info_hk_po.ccbins_id = orgName
alarm_info_hk_po.eqmt_id = deviceName
alarm_info_hk_po.bsn_cgycd = alarmType
alarm_info_hk_po.vlt_err_alrm_inf = vlt_err_alrm_inf
alarm_info_hk_po.alarm_level = alarmLevel
alarm_info_hk_po.stdt_tm = datetime.strptime(alarmTime,
'%Y%m%d %H:%M:%S')
if remark and len(remark) > 0:
try:
cur_date = time.strftime('%Y%m%d', time.localtime(time.time()))
folder_path = yaml_cfg.get(
'illegal_image_path_hk') + "/" + cur_date + '/' + orgName
if not os.path.exists(folder_path):
os.makedirs(folder_path)
img_path = (folder_path + "/" + alarm_info_hk_po.id + ".jpg")
cv.imwrite('D:/image/ab.jpg', img)
# f.save(img_path)
alarm_info_hk_po.uploadfiletrgtrfullnm = img_path
except Exception:
logger.error("存储hk告警图片失败, " + alarmType + "," +
vlt_err_alrm_inf + "," + alarmTime)
alarm_hk_dao.add_record(alarm_info_hk_po)
return make_response(status=HttpStatus.SUCCESS,
data={"code": "success"})