前言
本项目为IOT实验室人员签到考勤设计,系统实现功能:
◦人员人脸识别并完成签到/签退
◦考勤时间计算
◦保存考勤数据为CSV格式(Excel表格)
PS:本系统2D人脸识别,节约了繁琐的人脸识别训练部分,简洁快捷
该项目为测试版,正式版会加入更多的功能,持续更新中… 测试版项目地址我会放到结尾
项目效果图
登陆界面
主界面展示图:
签到功能展示
签退功能展示
后台签到数据记录
是否签到/退判断
项目环境
核心环境:
◦OpenCV-Python 4.5.5.64 ◦face_recognition 1.30 ◦face_recognition_model 0.3.0 ◦dlib 19.23.1
UI窗体界面:
◦PyQt5 5.15.4 ◦pyqt5-plugins 5.15.4.2.2 ◦PyQt5-Qt5 5.15.2 ◦PyQt5-sip 12.10.1 ◦pyqt5-tools 5.15.4.3.2
编译器
Pycham 2021.1.3 **Python版本 3.9.12**
Anaconda
辅助开发QT-designer
项目配置
代码部分
核心代码
python学习交流Q群:906715085#### 「MainWindow.py」UI文件加载: class Ui_Dialog(QDialog): def __init__(self): super(Ui_Dialog, self).__init__() loadUi(\"mainwindow.ui\", self) ##加载QTUI文件 self.runButton.clicked.connect(self.runSlot) self._new_window = None self.Videocapture_ = None
摄像头调用:
def refreshAll(self): print(\"当前调用人俩检测摄像头编号(0为笔记本内置摄像头,1为USB外置摄像头):\") self.Videocapture_ = \"0\" 「OutWindow.py」获取当前系统时间 class Ui_OutputDialog(QDialog): def __init__(self): super(Ui_OutputDialog, self).__init__() loadUi(\"./outputwindow.ui\", self) ##加载输出窗体UI ##datetime 时间模块 now = QDate.currentDate() current_date = now.toString(\'ddd dd MMMM yyyy\') ##时间格式 current_time = datetime.datetime.now().strftime(\"%I:%M %p\") self.Date_Label.setText(current_date) self.Time_Label.setText(current_time) self.image = None
签到时间计算
def ElapseList(self,name): with open(\'Attendance.csv\', \"r\") as csv_file: csv_reader = csv.reader(csv_file, delimiter=\',\') line_count = 2 Time1 = datetime.datetime.now() Time2 = datetime.datetime.now() for row in csv_reader: for field in row: if field in row: if field == \'Clock In\': if row[0] == name: Time1 = (datetime.datetime.strptime(row[1], \'%y/%m/%d %H:%M:%S\')) self.TimeList1.append(Time1) if field == \'Clock Out\': if row[0] == name: Time2 = (datetime.datetime.strptime(row[1], \'%y/%m/%d %H:%M:%S\')) self.TimeList2.append(Time2)
人脸识别部分
python学习交流Q群:906715085#### ## 人脸识别部分 faces_cur_frame = face_recognition.face_locations(frame) encodes_cur_frame = face_recognition.face_encodings(frame, faces_cur_frame) for encodeFace, faceLoc in zip(encodes_cur_frame, faces_cur_frame): match = face_recognition.compare_faces(encode_list_known, encodeFace, tolerance=0.50) face_dis = face_recognition.face_distance(encode_list_known, encodeFace) name = \"unknown\" ##未知人脸识别为unknown best_match_index = np.argmin(face_dis) if match[best_match_index]: name = class_names[best_match_index].upper() y1, x2, y2, x1 = faceLoc cv2.rectangle(frame, (x1, y1), (x2, y2), (0, 255, 0), 2) cv2.rectangle(frame, (x1, y2 - 20), (x2, y2), (0, 255, 0), cv2.FILLED) cv2.putText(frame, name, (x1 + 6, y2 - 6), cv2.FONT_HERSHEY_COMPLEX, 0.5, (255, 255, 255), 1) mark_attendance(name) return frame
签到数据保存与判断
csv表格保存数据 def mark_attendance(name): \"\"\" :param name: 人脸识别部分 :return: \"\"\" if self.ClockInButton.isChecked(): self.ClockInButton.setEnabled(False) with open(\'Attendance.csv\', \'a\') as f: if (name != \'unknown\'): ##签到判断:是否为已经识别人脸 buttonReply = QMessageBox.question(self, \'欢迎 \' + name, \'开始签到\' , QMessageBox.Yes | QMessageBox.No, QMessageBox.No) if buttonReply == QMessageBox.Yes: date_time_string = datetime.datetime.now().strftime(\"%y/%m/%d %H:%M:%S\") f.writelines(f\'\\n{name},{date_time_string},Clock In\') self.ClockInButton.setChecked(False) self.NameLabel.setText(name) self.StatusLabel.setText(\'签到\') self.HoursLabel.setText(\'开始签到计时中\') self.MinLabel.setText(\'\') self.Time1 = datetime.datetime.now() self.ClockInButton.setEnabled(True) else: print(\'签到操作失败\') self.ClockInButton.setEnabled(True) elif self.ClockOutButton.isChecked(): self.ClockOutButton.setEnabled(False) with open(\'Attendance.csv\', \'a\') as f: if (name != \'unknown\'): buttonReply = QMessageBox.question(self, \'嗨呀 \' + name, \'确认签退?\', QMessageBox.Yes | QMessageBox.No, QMessageBox.No) if buttonReply == QMessageBox.Yes: date_time_string = datetime.datetime.now().strftime(\"%y/%m/%d %H:%M:%S\") f.writelines(f\'\\n{name},{date_time_string},Clock Out\') self.ClockOutButton.setChecked(False) self.NameLabel.setText(name) self.StatusLabel.setText(\'签退\') self.Time2 = datetime.datetime.now() self.ElapseList(name) self.TimeList2.append(datetime.datetime.now()) CheckInTime = self.TimeList1[-1] CheckOutTime = self.TimeList2[-1] self.ElapseHours = (CheckOutTime - CheckInTime) self.MinLabel.setText(\"{:.0f}\".format(abs(self.ElapseHours.total_seconds() / 60)%60) + \'m\') self.HoursLabel.setText(\"{:.0f}\".format(abs(self.ElapseHours.total_seconds() / 60**2)) + \'h\') self.ClockOutButton.setEnabled(True) else: print(\'签退操作失败\') self.ClockOutButton.setEnabled(True)
项目目录结构
后记
◦因为本系统没有进行人脸训练建立模型,系统误识别率较高,安全性较低
◦系统优化较差,摄像头捕捉帧数较低(8-9),后台占有高,CPU利用率较高
◦数据保存CSV格式,安全性较低
来源:https://www.cnblogs.com/1234567FENG/p/16376000.html
本站部分图文来源于网络,如有侵权请联系删除。