Inspiration

  • Make Sense of Data track use some application of data science or machine learning
  • Indeed Challenge innovative projects and solutions that tackle societal problems

What it does

Family members cannot get in contact with each other after natural disasters, and disaster victims might be hospitalized without identification. reconnect. bridges the gap: first responders upload images and general information about disaster victims, family members search for their loved ones by uploading photos of them, our facial-recognition system returns top 10 most similar profiles, and the user selects the correct match.

How we built it

Deep Learning for Facial Matching

  • Step 1: Transform images to 2622 dimensional vectors
  • Step 2: Deep Learning Training and Inference
  • Step 3: Calculate vector similarity (cosine distance)
  • Step 4: Verify Matches (Match = Distance less than threshold)
  • Step 5: Return results (top 10 most similar profiles)

Build up the web application

  • Step 1: Upload Image
  • Step 2: Run the facial matching algorithm back-end
  • Step 3: Return top-10 most similar people
  • Step 4: Fetch information and give feed back

What's next for Reconnect

Area for improvement

  • More accurate facial recognition- we recognize that rapid advancements in technology may produce better infrastructure in the near future

Future implementations

  • Ability to be notified of new matches ** General setup: store user-input images in a separate database ⇒ run matching algorithm for newly uploaded profiles ⇒ email/phone alert when new matches are found

  • Age-adjusting algorithm- find matches using an older image of the person (when they were younger)

Built With

Share this project:

Updates