Due to the COVID-19 pandemic, many schools and universities have decided to take tests and classes online. Students taking tests online tend to cheat during the exam and it may be a short term solution for the students now but in the long term, it won't help. To solve this problem for my school we made this application.

What it does

When a student starts the test he/she can start recording his screen and webcam until he finishes the test. Once the test is over he/she can upload it to their respective teacher viz their student's dashboard. Once the upload is done the machine-learning model flags the students who it thinks may have cheated while taking the tests and flags each student. These data can be viewed by the teacher for up to 24 hours and after that will deleted.

How I built it

It monitors the students in every way possible without much computer resources! it takes in 3 inputes: 1) Webcam video 2) Mic Audio 3) Screen audio and screen capture with these three inputs, an image classification model differentiates between different screens. For example, the AI declares the students are not copying if the screen capture does not have any audio playing or if the student stays within Google Classroom, Google Forms and Google meet(These are the only screens I used to train it as these are the three applications used by my school for testing). It also uses face tracking and audio recognition for webcam and MIC to determine if the student is referring to some other books or is seeking help. It can also be used for student verification. It also keeps track of switching tabs using JavaScript.

Challenges I ran into

One of the challenges we faced was not everyone has a good internet connection for live proctoring(at least in my school). We overcame this problem by making the application work with slow internet speeds. One of the other challenges we're currently trying to overcome is the machine-learning model problem. Basically the image classifier takes a lot of computing power and the machine starts to slow down, this we're trying to solve by using a cloud-based model and not using in device intelligence.

Accomplishments that I'm proud of

The main accomplishment that we are proud of is that we have running a web-based prototype that is able to run in a live demo at the final presentation.

What I learned

Not only did we have to overcome different challenges but we also learned some lessons on the way. It showed us once again the need for constant re-evaluation of the requirements, strong teamwork, and adaptability.

What's next for Equidor

To turn our project into a final project different steps are needed. From an improved software architecture and performance, a fitting billing model to enhanced usability. We are also looking into ways for advanced live proctoring

Share this project: