Automatic Visual Inspection

IoT

Background

The client runs a smart screen advertising business. Ad screens, located across the city and in various shops, were displaying contextual commercials, based on gender, age, and other attributes. The company uses raw WiFi signals data from custom access points distributed throughout numerous city locations for further analysis, feature extraction, and to build a scalable and fault-tolerant system.

The client needed a highly scalable platform to handle big data from numerous on-premise sensors embedded in displays and its integration with the existing system.

Detecting and categorizing customer visits into three main groups:

  • people who are quick buyers,
  • people who are buyers and spend much time in a shop,
  • people who pass by a shop.

Our work

We managed to perform some sophisticated feature engineering and create a distributed, highly scalable model consisting of Self Organizing Maps and K-Means. We also implemented and integrated a fault-tolerant platform on the cloud to gather data from numerous sensors.

The outcome

The primary users of our solution were the shop owners who could see the statistics on their clients and how much time they spent shopping. Thanks to our solution, the shop owners can adjust the size of their staff and optimize their budgets.

Technology used
  • Google Cloud
  • Scala
  • Java
  • Spark