TransitPay | Devpost

Inspiration

Currently, we need to buy a physical or a ticket on our mobile devices and redeem them as and when required, which is just one more trivial thing taking the space in our heads apart from much more important things that keeps us up with the kardashians.

We have devised a solution to this significant travelling problem, where you never even have to remember to buy the tickets.

What it does

TransitPay is an application that tracks your journey in the train and automatically charges you according to your itinerary. No need to buy or redeem the tickets, just carry your phone with you and you are all set to board!!

How we built it

TransitPay system aims at reducing the hassle of the whole ticketing system. The user can simple install the app into their mobile devices and load a certain amount of money into their wallets. The train doors are equipped with NFC radars that the application will detect and send a checking ping to the server once a user enters the train, this will check the user into the train and the journey will begin.

The same sensors will detect the user moving out of the train and the application will check him out of the train. Once checked out, the application will run the calculation on the user’s journey and calculate the fare, which will then automatically be deducted from the user’s wallet.

TransitPay is not just an application, it's a whole system, we have cameras installed on all out exits that count the number of people entering and exiting using a Computer Vision model, this model keeps a count of all the activities at the doors and sends the count of total people that are supposed to be in the compartment.

This number is then compared with the count of the live tickets in that train, in case of any disparity between the numbers, the conductor is informed and they can check the particular compartments.

Challenges we ran into

Integrating the application (android app) with the server and the CV model was more complicated that it seemed to be.

Accomplishments that we're proud of

Finally, integrating the android app with the server and model. XD

What we learned

We developed new found love and hate with the following technologies:

OpenCV - Love it
Caffe - Hate it (Made me cry)
Node JS - Love it
Android Studio - We hate that we love it. ;)

What's next for TransitPay

The current iteration of TransitPay works on a generlised MobileNet model that detects 9 classes other than Humans. Although this is working on the shown use cases, it can be heavily improved with a proper custom data set to train the model. By custom, we mean videos and images of humans from a top view. We tried annotating a few publically available videos, but we did not have the time to complete and use it. There is a video attached in the files, which shows the current systems' (CV system), performance in a very crowded setting. This performance is something that can easily improved with custom model training and tuning.