Inspiration

Nowadays, almost every retailer has security cameras. They are, however, missing out on a significant potential because no insights are generated from them. Our Approach Room Vision assists retailers in obtaining insightful data about on-site stores utilising CCTV footage and applying it to data-driven store optimisation.

What it does

RoomVision can help get the trajectory of clients in a room, enabling retailers to study the behaviour of customers and redesign their store layouts. We can also get the number of users in a given region for certain timestamps, which can help in studying the engagement time, queueing time, times per section, Conversion rate. We offer daily statistics and heat-maps about people's interactions in the store.

Why our solution matters

Room Vision assists merchants on numerous levels. First, the solution improves the visitor experience. It ensures that the marketing team's efforts result in high-quality retail traffic. By evaluating in-store conversions and maximising store revenues, operations can increase in-store conversions and maximise store revenues.

How we built it

To build RoomVision, we used the latest cutting edge technologies, including machine learning and cloud computing. We used TensorFlow to detect people's positions and their movements, combined with our trained model built with Azure Custom Vision. We also used Azure SQL Database to store data gathered with end devices like IP Cameras or IoT devices. These data are provided and retrieved using Azure Functions, which provides a fine control over hardware usage and system scalability. Architecture

Challenges we ran into

  • Link external camera with web apps.
  • Technical problems due to network protocol restrictions, specifically RTSP (Real Time Streaming Protocol) which couldn't be used and therefore limited our potential technical solutions.
  • Hybrid team (2 in-person and 1 remote) but we worked that out.

Accomplishments that we're proud of

Creating a minimum viable product that solves a real world problem by helping store owners in making smart decisions.

What we learned

  • Azure SQL Database & SQL Server Management Studio for creating, deploying and managing the database.
  • Azure Custom Vision for training a model that detects masks in photos.
  • Azure Functions as an HTTP trigger to communicate with the database.
  • ASP .NET used to implement the functions in Azure Functions.

What's next for RoomVision

  • Implement more advance analysis tools based on machine learning.
  • Market validation by trying our prototype in real world scenarios.

Built With

  • azure
  • azure-custom-vision
  • azure-functions
  • azure-sql-database
  • azure-static-web-apps
  • chart.js
  • heatmap.js
  • react
  • react-material-ui
  • tensorflow.js
Share this project:

Updates