Inspiration

Nowadays, large corporations are spending more and more money nowadays on digital media advertising, but their data collection tools have not been improving at the same rate. Nike spent over $3.03 billion on advertising alone in 2014, which amounted to approximately $100 per second, yet they only received a marginal increase in profits that year. This is where Scout comes in.

What it does

Scout uses a webcam to capture facial feature data about the user. It sends this data through a facial recognition engine in Microsoft Azure's Cognitive Services to determine demographics information, such as gender and age. It also captures facial expressions throughout an Internet browsing session, say a video commercial, and applies sentiment analysis machine learning algorithms to instantaneously determine the user's emotional state at any given point during the video. This is also done through Microsoft Azure's Cognitive Services. Content publishers can then aggregate this data and analyze it later to determine which creatives were positive and which creatives generated a negative sentiment.

Scout follows an opt-in philosophy, so users must actively turn on the webcam to be a subject in Scout. We highly encourage content publishers to incentivize users to participate in Scout (something like $100/second) so that both parties can benefit from this platform.

We also take privacy very seriously! That is why photos taken through the webcam by Scout are not persisted anywhere and we do not collect any personal user information.

How we built it

The platform is built on top of a Flask server hosted on an Ubuntu 16.04 instance in Azure's Virtual Machines service. We use nginx, uWSGI, and supervisord to run and maintain our web application. The front-end is built with Google's Materialize UI and we use Plotly for complex analytics visualization.

The facial recognition and sentiment analysis intelligence modules are from Azure's Cognitive Services suite, and we use Azure's SQL Server to persist aggregated data. We also have an Azure Chatbot Service for data analysts to quickly see insights.

Challenges we ran into

CORS CORS CORS!. Cross-Origin Resource Sharing was a huge pain in the head for us. We divided the project into three main components: the Flask backend, the UI/UX visualization, and the webcam photo collection+analysis. We each developed our modules independently of each other, but when we tried to integrate them together, we ran into a huge number of CORS issues with the REST API endpoints that were on our Flask server. We were able to resolve this with a couple of extra libraries but definitely a challenge figuring out where these errors were coming from.

SSL was another issue we ran into. In 2015, Google released a new WebRTC Policy that prevented webcam's from being accessed on insecure (HTTP) sites in Chrome, with the exception of localhost. This forced us to use OpenSSL to generate self-signed certificates and reconfigure our nginx routes to serve our site over HTTPS. As one can imagine, this caused havoc for our testing suites and our original endpoints. It forced us to resift through most of the code we had already written to accommodate this change in protocol. We don't like implementing HTTPS, and neither does Flask apparently. On top of our code, we had to reconfigure the firewalls on our servers which only added more time wasted in this short hackathon.

Accomplishments that we're proud of

We were able to multi-process our consumer application to handle the massive amount of data we were sending back to the server (2 photos taken by the webcam each second, each photo is relatively high quality and high memory).

We were also able to get our chat bot to communicate with our REST endpoints on our Flask server, so any metric in our web portal is also accessible in Messenger, Skype, Kik, or whatever messaging platform you prefer. This allows marketing analysts who are frequently on the road to easily review the emotional data on Scout's platform.

What we learned

When you stack cups, start with a 3x3 base and stack them in inverted directions.

What's next for Scout

You tell us! Please feel free to contact us with your ideas, questions, comments, and concerns!

Built With

Share this project:
×

Updates