I-Corps: Automated Attendance Check by Using Smartphone Cameras

在教室等场景中检查出勤情况通常需要教师通过阅读花名册上的名字或要求学生签署出勤表来逐一识别每个学生。然而，这种传统的方法面临着两个问题：当学生人数较多时，朗读学生的名字可能会占用几分钟的讲课时间；让学生签署考勤单容易被欺骗，因为他们可以在课堂上签名自己的名字和缺席的同学的名字；教师通过手动查看每一张考勤表来计算每个学生一学期的总出勤率并不是一项理想的任务。这个i-Corps团队提出了一种高效而准确的方法来完成这项任务。通过使用智能手机摄像头拍摄教室中的学生人脸，该团队提出了一个统一的视觉人脸检测、跟踪和识别算法框架，以同时识别视频中的多个人脸。该系统包括以下步骤：教师在自己的智能手机上安装所提出的App；在第一节课上，教师使用智能手机摄像头拍摄课堂上的学生人脸的短时间视频。该应用程序将自动为课程建立人脸数据集，教师只需识别第一节课的人脸数据集；在其余课程中，教师拍摄每节课的视频，应用程序将进行自动出勤检查。拟议的智能手机应用程序将执行多对象跟踪，将检测到的人脸(包括假阳性)关联到人脸跟踪程序(每个跟踪程序包含同一个人的多个实例，具有不同的姿势、光照等)。然后将每个人脸轨迹中的人脸实例聚类成少量的簇，以较少的冗余度实现稀疏的人脸表示。

Checking attendance in scenarios such as classrooms commonly needs an instructor to recognize each student one by one by reading the names on a roster or ask students to sign up the attendance sheet. However, this traditional method faces two problems: reading students' names may occupy minutes of lecture time when the number of students is large and letting students to sign up an attendance sheet is prone to be cheated since they can sign their own names and their classmates' names who are absent in the class; it is not a desirable task for instructors to calculate the total attendance of every student in a semester by going through every attendance sheet manually. This I-Corps team proposes an efficient and accurate way to accomplish this task. By taking videos of student faces in classrooms using Smartphone cameras, the team proposes a unified framework of visual face detection, tracking and recognition algorithms to recognize multi-faces in the video simultaneously.The proposed system has the following steps: instructors install the proposed App on their own Smartphones; in the first class, instructors use the Smartphone cameras to take a short-period video of student faces in the classroom. The application will automatically build a face dataset for the course and the instructor only needs to identify them for the first class; in the remaining classes, instructors take videos of each class and the application will do automated attendance check. The proposed Smartphone App will perform multi-object tracking to associate detected faces (including false positives) into face tracklets (each tracklet contains multiple instances of the same individual with variations in pose, illumination etc.) and then the face instances in each face tracklet are clustered into a small number of clusters, achieving sparse face representation with less redundancy.