A prevalent issue within migrant and refugee families is the separation of members over the course of a journey. We often hear stories of family members (siblings, parents, grandparents, etc.) who reunite decades after a tragedy occurs. With the right technology in place, we believe that if people get separated from each other, they won't have to wait years to reunite and restart their lives in a new place.
What it does
Our multi-platform application allows users to input pictures of their missing relatives and potentially receive information about them. The software leverages current facial recognition technology to match an inputed image to the many images of missing people/refugees that exist in NGO databases. The search returns the ten images with the best similarity score.
How we built it
We structured our work into three areas: (1) dataset collection, (2) machine learning, and (3) web, bot, and SMS access points.
(1) Dataset collection. We scraped the International Committee of the Red Cross's "Trace the Face" database of photos of migrants who had been separated from their families. This dataset spanned almost 4,000 images, compiled from Red Cross societies in 28 European countries.
(2) Machine learning. We used the Microsoft Cognitive Services Face Recognition API to train a face similarity model. We also constructed Python APIs for mediating our custom interactions with the Face Recognition service. Our APIs allow us to easily upload new facial photos to the Microsoft portal and then re-train the model.
(3) We built user-friendly access points to the face recognition service via web app, messenger chatbot, and SMS bot.
Challenges we ran into
A lot with the web side.
Accomplishments that we're proud of
1) Gathering a large and relatively clean database, especially considering the decentralized nature of missing persons databases 2) Successfully leveraged Facial recognition APIs for reasonable photo matches 3) Onboarding of chatbot onto Facebook messenger 4) Developing a back-end service that multiple interfaces could extract data from 5) Developing a cleaned up web platform for ease of use
What we learned
1) Learned more about multi-platform integration 2) Learned about facial recognition
What's next for Vinculum
1) Develop the accuracy of the facial recognition models 2) Gather more data for 1) 3) Expand into areas concentrated with refugees (such as UN camps) so families can start finding each other 4) Leverage technology in other spaces (such as human trafficking)