Customer Case Study
TensorIoT utilized AWS to build a machine learning computer vision solution that was capable of recognizing, recording, and counting casino chips to determine what bets were being placed at a specific gaming table.
Casinos and other gambling entities have spent decades tracking the people moving within gaming spaces, using constant monitoring to prevent cheating and theft. However, despite an existing elaborate camera system with constant surveillance, most of the Gaming industry doesn’t possess good actionable data on individual gamblers that isn’t reliant on some form of rewards card. Using rewards cards relies on the customer to manually insert their card into a machine or show the card at a table each time they play, which constantly reminds the customer that they’re being tracked and adds an extra layer of interaction between the casino and the customer. [CLIENT] wanted an introduction into the world of leveraging IOT to improve customer tracking and rewards programs, and TensorIoT was ready and able to take them to the next level of business. [CLIENT] needed a robust system with the ability to collect player data and compile it in a central place where all the systems for a casino can reference client data. The first step in the process was designing a method for Computer Vision technology to accurately identify and count the number of casino chips.
Counting casino chips is a new and novel task for computer vision technology. In order to properly train the ML technology to count chips, there were very specific constraints on camera choice and position. Chips are normally read via bar codes which made it difficult to identify chips based on images. If a recording camera was bumped or changed, then it altered the ability of the machine to identify chips. It was also a challenge to design the model for handling chip recognition since gamblers don’t always place chips in the same precise location on the gaming table.
Chip recognition was a major issue that required a combination of machine learning models to address. To gain a thorough understanding of the task at hand, TensorIoT set up a Baccarat table with the same layout and cameras as a live casino environment, then used this setup for image processing and testing using cameras to train the ML models in critical areas like zone detection. Organizing and tracking iterations of the required training the ML model trainings necessary to achieve the task chip recognition and counting were made easier by employing the AWS Sagemaker service. AWS Sagemaker trained models were then easily deployed the edge (casino) environment using AWS Greengrass’s ML resource capability. Using live modeling of the gaming table, TensorIoT developed a sequence of different machine learning models that relied on each other to correctly to preprocess images and apply them to models to come up with the chip total. Proper tabulation of chips enables bet tracking and monitoring without requiring human intervention.