Student attendance system in classrooms: The idea of this code is to verify is the subjects that are enrolled are present in the session images. In this simple example, there are 27 students in 5 sessions. The result is shown in result.png.
Results of our Student Attendance System [1]. In this example of 25 sessions, the selected student was not present in 5 sessions (see black squares in the right image), i.e, the attendance for him is 20/25 = 80.0%. In the middle image, the session of March 29th is shown, where the student was recognized in the last row of the classroom (see zoom of the red square in the left panel). In these experiments, the images were acquired using a smartphone camera.
- Install Homebrew /usr/bin/ruby -e "$(curl -fsSL https://raw.githubusercontent.com/Homebrew/install/master/install)"
- Re-install boost-python: brew reinstall boost-python --with-python3 --without-python
- pip3 install --upgrade pip
- pip3 install cmake
- cd dlib
- python3 setup.py install
- pip3 install face_recognition
- pip3 install scikit-image
- pip3 install opencv-python
- pip3 install keras
- pip3 install tensorflow
- git clone https://github.com/davisking/dlib.git
- git clone https://github.com/ageitgey/face_recognition
- git clone https://github.com/nyoki-mtl/keras-facenet
- in folder keras-facenet/model create keras
- in folder keras-facenet/model/keras copy this model https://www.dropbox.com/s/o65ow1cl3lcwd1o/facenet_keras.h5?dl=0
In a folder called "facer-classroom" copy the following files:
- main.py
- utils.py
- data/enroll
- data/sessions
Run main.py (Expected output: attendance report and face images of each student in every session).
- If you want to understand how an enrolled student is recognized in a session image, it is recommended to see the code of facer-simple
- To understand main.py: edit the file and read the comments, all functions are defined in utils.py
- The demo uses FaceNet. If you want to run the demo using dlib see inputs in next section.
- In line 80 of utils.py please include your path for facenet_keras.h5
Check definitions in lines 12-29 of main.py:
- for definition of face detection method see variable fd_method
- for definition of face recognition method see variable fr_method
- for extraction of descriptors of enrolled face images set first element of variable extract_desc
- for extraction of descriptors of session face images set second element of variable extract_desc
- for display result image set variable show_img
- id of subjects are given in student_list.csv
- enrolled images are in data/enroll
- session images are in data/sessions
Check the parameters of the demo code:
- FIRST TIME:
- to extract and save the features of sessions and enrolled subjects, set in definition parameters of main.py: extract_desc = [1, 1]
... it takes 5-10 minutes to extract and save the descriptors. After this step, files .npy are created in sessions directory and enrollment directory.
- SECOND TIME:
- when features are already extracted. Thus, you can , set in definition parameters of main.py: extract_desc = [0, 0]
The obtained results are the same results obtained in FIRST TIME.
- THIRD TIME: If you want to obtain only the attendance record without displaying the detected faces, , set in definition parameters of main.py: show_img = 0
This is very fast!
[facer] : ----------------------------------------------------------------
[facer] : fd = HOG, fr = FaceNet, sc = CosSim(uninorm=1), th = 0.45
[facer] : loading face recognition model FaceNet...
[facer] : loading descriptors from data/enroll/000297/FaceNet...
:
[facer] : >>session 0/5...`
:
[facer] : assistance report in sessions 1...5
297 - Name Last_Name_20 : 5/5 = 100.00%
:
326 - Name Last_Name_10 : 4/5 = 80.00%
References:
- [1] Mery, D.; Mackenney, I.; and Villalobos, E. Student Attendance System in Crowded Classrooms using a Smartphone Camera. In 2019 IEEE Winter Conference on Applications of Computer Vision (WACV2017), 2019.
Note: The students of these pictures gave their permission to be part of this research. The pictures can be used free of charge, for research and educational purposes only. Redistribution and commercial use is prohibited.