Pehechaan

AI enabled ID verification; Based on OCR and other Computer Vision techniques.

Comment

Inspiration

I've noticed that many companies like Binance, Facebook and Zerodha require their customers to verify their ID's with a video, or a photo. All of these companies, specially Facebook and not so much Zerodha are known for breaching user privacy.

Pehechaan aims to eliminate the need of humans for ID verification completely. More on how it works in the next section.

What it does

We use OCR techniques to read from ID proof, use liveness detection methods such as a blink test and by using pre-trained models, we will be able to verify if the person is real or not. Moving further we compare your photo (i.e. the person who is being verified, with the photo on the ID provided. Upon getting a satisfactory score, the person who is being vetted is allowed to move ahead, else they're held back.

Being an AI model, there is always room for mistakes, error and false positives/negatives. This can be improved with training on larger datasets.

How we built it

The project's back end is heavily dependent on PyTorch. I've used open-cv, image AI and face_recognition libraries to process images, detect people in ID card, and compare faces respectively. The project uses the Yolov3 model to recognize faces of people in images. The goal is to completely remove human intervention in ID verification for privacy reasons.

Currently, Pehechaan can:

  1. Detect a person in an ID card
  2. Read/Process information off the ID card
  3. Take images of the person being vetted, and compare it to the image on the ID card
  4. Conduct a blink test to verify if the person is real or not

Challenges we ran into

Compatibility issues.... My project worked fine on Windows, until I had to install the face_recognition library. Dlib compilation took me 30 minutes on my raspberry pi, not to forget I had issues with setting up a camera with my raspberry pi.

Accomplishments that we're proud of

I'm proud of that fact that my idea came turned into reality, maybe not completely, but most of it. Most ideas just remains ideas and mine didn't.

What we learned

I learnt the different libraries for OCR, I was aware of the OG tesseract until I started researching OCR on the internet. I also learnt how to compile dlib and windows' dependency issues for multiple libraries.

What's next for Pehechaan

For now Pehechaan is a raw script, just a cluster of functions. I plan on building a web interface with a Flask backend, I also plan on adding a sqlite connector so it is easier for anyone using this tech to query SLQ databases. A proper front end is also planned. Since I learnt about this hackathon 2 days prior to the submission date, this is the best I could come up with!

Built With

  • ocr
  • open-cv
Share this project:

Updates