EyeDetectionWebsite-Ingress
Use machine learning in JavaScript to detect eye movements and build eye-tracker experiences! and this tool relive the sever by performing in frontend
Created on 14th February 2021
•
EyeDetectionWebsite-Ingress
Use machine learning in JavaScript to detect eye movements and build eye-tracker experiences! and this tool relive the sever by performing in frontend
The problem EyeDetectionWebsite-Ingress solves
The problem
As we all know due to covid, the exams for all the student are done online, there are many website that are responsible for conducting exams for an institute one of them is
WheelBox,
There are 1000s of hiring test that that are conducted on wheelbox, wheelbox this year revenue was 2.7 million dollars,
Still Wheelbox has a lot of flaws,
I want to take your attention towars how wheelbox works, wheelbox continuously(in every 6sec) or so send 1 image per pc to the backend and then that image is being
parse through ai models and then ai models gives an output that whether student is on the screen or on off the screen ie is he doing cheating or not
Now there are 3 layers
first taking snap from dom tree(perform in frontend on the user pc) convert that image into an url
Here is an example of the url
<A big URL converted from image to a bob of url almost 2000 character long>
this image is then sent to backend(almost 0.4 mb file(estimated by doenloading the file))
now this 0.4 mb image is sent to backend and then the image is passed to an ai model which check whether if the file has some problems or not
If problem send warning otherwise continue
Now think for one person
If the test is of 1 hour
in one min 10 images is sent to backend and in 1 hour 600 images is sent by one user
Now if we calculate it in mb then 600*0.4 =240 mb file has been sent and then its been check by ai
and this is only the image part what about the exam
almost half a gb is required for one user
now think about it if there are twenty thousand student then almost ten thousand gb od data is needed to be sent within 1 hour and after 1 hour the server will be useless
and one need to scale down the server manually
10000 gb is a lot of data, which burdened the whole website and can possible make site down
Our aim is to make all this file that require a backend convert it into frontend, what will be the advantage
- Smooth paper conduction,
- server will not get down easily
Challenges we ran into
During the start of the project when our team was trying to think which technology we should use, We waste alot of our time fighting over python or javascript. but atlast we all decided to work with javascript after the project initial stage was done i asked all my members to ckeck whether the problem is running or not and there it was not working on any single person, so for that we faces some issues there(the stack oberflow thread that resolve issue for all of us is https://stackoverflow.com/questions/64573177/unable-to-resolve-dependency-tree-error-when-installing-npm-packages )
At night I face some issues during the creation of machine larning model the maths was just too complex I had to manually measure 50 test cases and by taking the average i come to a solution for the maths which took a ridiculous amount of time,
After that done we face some issue on setting the cideo tag in the website, which was resolved by simply using video tag.
After that there were some issues we face during the creation of website like how to make it responsive and where to put the video tag
In last we face issues regarding creation of tutorial video and some nitpicks issues was resloved in the end
Overall It was quite challenging for my team but I am happy we pull through and complete the project that we previously anticapated
Technologies used