Using Classification for Automatic Attendance Management

1. Learning Phase - In this phase, the software system can be trained to recognise the faces of the students through the one time input of facial data. Here, machine learning will occur in an unknown environment.
2. Monitoring Phase - The next phase will use classification. Here, the scanner (here a camera), will capture facial inputs of students who are entering and exiting. The software will check each face and will try to match it with the existing set of facial data. If the matching of features is above a certain percentage, the person's attendance will be recorded.
3. Real-time detecting phase - This phase is an implementation of the monitoring phase for broader performance analysis. In this, through machine learning, not only the attendance of students but also their stay in classes can be recorded. This will help in minimising “bunk” culture and will also generate a report. ^[3]

Input Dataset:

As the automated system uses facial recognition for attendance registering, there can be three broad types of data which should be the input. These data sets become input at two stages. The first stage is Machine Learning stage, and the second stage is Real-time implementation.

• For the machine learning stage, the algorithm of the system can be trained using data sets of nearly a million or more different images. Using Labeled Faces in the Wild or LFW dataset, which has more than 13,000 images of faces were fed as input and using various methods, the machine learning can reach or even exceed an accuracy of 99.3%.
• In the Machine Learning Phase, when the system reaches the desired accuracy, images of students which attendance is to registered becomes the data set. This is termed as the primary data set and is used for referencing every time a person comes across the system.
• The third stage input is the real input of the system. This will be the data captured by the scanner part of the system. These input data will be just images, from which facial features can be extracted and matched with the existing primary dataset. ^[4]

The data collection of the third stage input can be done by cameras, which can be two in number. Here the cameras can act as a scanner. As soon as the live stream is retrieved, the video frame is captured and is converted into grayscale. This grayscale feed is then fed into a feature extractor which starts detecting facial features, essentially faces. The faces are then aligned as per system and encoded. This process creates input data. The facial data thus collected is then sent to a matcher. ^[5]

Class Label Outputs

The input data goes into the Matching module. Here, each encoded faces constitutes the input data set. Here comes the use of classification. Each of these input data sets acts as a discrete value, and a matching process starts. This process starts matching the discrete input value with the much larger primary set database of students and then each face from the database is matched with the input set to predicts which student is it, or is it not. ^[6]

If the Detection ID is confirmed, at first camera, which is at the ENTRY side of the classroom, in inward direction then the output of the system comes as YES. At output YES, the ENTRY is recorded in a report. If the ID isn't confirmed, the output is NO, and the frame is skipped, and the next frame is brought on to record more.

If the Detection ID is confirmed at the second camera, which is at the EXIT side of the classroom in an outward direction, the output of the system comes as YES, and the EXIT is recorded. If the ID is not confirmed, then the output of the system becomes NO. Here, the action which follows this output is skipping frame, and the process continues.

Since, both the entry and exit of each of the student is being recorded, the following decisions can be taken:

• Recording Attendance of each student at ENTRY record at first camera.
• Demolishing bunk culture by recording time between ENTRY and EXIT record. This can be reported, and the performance evaluation can detect bunk.
• The same system’s output can be optimised for industries and various organisations to record time of working of each personnel.

Input Features Example

The input features include -

• Facial boundaries
• Facial Structure
• Nose
• Distance between two eyes
• Mouth
• Chin

These above input features combined together will form the complete input facial feature set.

The output classes and values will be:

• Face Match confirmation of each student is itself a class
• ENTRY and EXIT records

Sample Size

The classroom used here as an example of working of this solution has 100 students. First, images and facial features of each of these 100 students is made the primary data set. Since the attendance recording will be carried out through two different cameras, the number of samples of input can vary from 1 face to up to 10 faces all in the range of the camera. Every time, the camera will tend to record faces, and total sample size at one instance of classroom entry or exit should be 100 for a class of 100 students. For this, the students will have to face the camera for entering and exiting the classroom. The number of instances too will vary from 6 on both of the cameras (owing to 6 different classes) to more of students keep moving out or in as per will. For a class of 100 students, these instances can even go up to 1000 times or even more.

Similar research works

Many similar attendance registering systems are either in the research phase or are already being used all over the world. Some of the prominent ones are:

• Many similar systems use Multi-Task convolutional neural network face recognition algorithms, which is much better and complex than what the solution is prescribed in this paper. These systems are more accurate and are generally used in the purpose of security of important places. ^[7]
• Using Fingerprint based systems for Attendance Management: Many research work has been conducted on this, and through machine learning classification, the Fingerprint matching can be done from the existing set of fingerprints recorded earlier. This system is used in various organisations, institutions, and companies for staff members. Instead of signing an entry sheet, all they have to do is to scan their thumb. ^[8]

Though there are various works being carried out in using machine learning based face recognition systems, all are security-based and hence use a much complex facial recognition algorithms. These use colour images for better detection. But, when we use it for a school attendance registering system, it won't be fruitful. If the same process of used as discussed above, the cost will be drastically low and the process will be much more effective than that of Fingerprint recognition systems.

References:

1. Wang, D., Fu, R. and Luo, Z., 2017. Classroom Attendance Auto-management Based on Deep Learning.
2. Cunningham, P. and Delany, S.J., 2007. k-Nearest neighbour classifiers. Multiple Classifier Systems, 34(8), pp.1-17.
3. Khan, M. and Gopale, A., 2017. Automated attendance management system using face recognition.
4. Chintalapati, S. and Raghunadh, M.V., 2013, December. Automated attendance management system based on face recognition algorithms. In 2013 IEEE International Conference on Computational Intelligence and Computing Research (pp. 1-5). IEEE.
5. Jain, A.K. and Li, S.Z., 2011. Handbook of face recognition. New York: springer.
6. Balcoh, N.K., Yousaf, M.H., Ahmad, W. and Baig, M.I., 2012. Algorithm for efficient attendance management: Face recognition based approach. International Journal of Computer Science Issues (IJCSI), 9(4), p.146.
7. Zhang, C. and Zhang, Z., 2014, March. Improving multiview face detection with multi-task deep convolutional neural networks. In IEEE Winter Conference on Applications of Computer Vision (pp. 1036-1041). IEEE.
8. Saraswat, C. and Kumar, A., 2010. An efficient automatic attendance system using fingerprint verification technique. International Journal on Computer Science and Engineering, 2(02), pp.264-269.