I want to create a facial-attendance system for my school of about 2000 students. A camera will be placed in the entrance of the school . Students will pass looking at the camera one by one. Every student will face the camera for about 1.5 seconds. The camera will be a 720p camera of 30fps. The camera will feed the input to the backend and the backend powered by python will analyze the input and mark the attendance in a DB. As I am a beginner to ML. I don’t know which technology/model to use to recognize the face in a prepared model of about 2000 students in a short time with high accuracy as I have to feedback to the student about the successful attendance. On every annual year a human will enter the details of each new comer with the video of about 5-15 seconds of each student into the model. The training time of model can vary from hours to days. Any help/suggestion is highly appreciated
I’m very underqualified to recommend any current technology or methods. However, assuming you intend to put this into production, have you considered the ethics and consequences of doing so?
I am not planning to put this into production anytime soon as this is for demonstration only and the images are not gonna be stored anywhere. After the generation of the model Images will be deleted and I don’t think it is possible to scrape any useful image from from the generated model.
There are pre-trained models available that can help you to perform the task.Although I cant explain here how you can recognize faces, but you can use Opencv library to do your task.You ll need two addional libraries called dlib and face_recognition
The technique generally used is called deep metric learning.