Azure Percept is an end-to-end rapid-prototyping solution for deploying AI models on Edge
Just like any other tech hub, our workplace runs on coffee, but often times when I want to go make a cup of coffee, there is queue at the machine. Since I tend to have only few mins between meetings or coding sessions to grab coffee, waiting in line is not ideal.
Tricky problem, right? *My developer brain thinks of a fun project*
I built an office coffee bot with Azure Percept using in-built people detection AI model that can tell me when the coffee machine is unoccupied. AI + coffee - fun combo, isn’t it? Now, let’s get to details.
Azure Percept CoffeeBot Setup
The development kit comes with AI models that include people detection, object classification, etc., and can be easily integrated with Azure services.
Using Azure Percept studio, you can manage your devices and deploy custom models built with Azure Custom Vision. I decided to deploy the people detection model on Percept for my use case and configured it to continuously stream detection data onto Azure IoT Hub. On IoT hub, you can configure a device to receive these events.
Next, I plugged Stream Analytics to IoT Hub to in order to run queries on streaming data. The input of the stream analytics job is set as IoT Hub device I had created earlier. The output of the job is sent to CosmosDB in this usecase.
Stream Analytics job query counting number of people detected every 3 minutes
The data is stored as CosmosDB documents
In order to finalise my bot, I setup a Microsoft Teams bot that calls an Azure Function App via HTTP trigger in order to instantly answer whether the coffee machine is free.
This is how the whole setup looks -
The people counter data in Cosmos DB can also be visualised over a period of time using PowerBI or similar tool to know when people are most at the coffee machine which can lead to other automation possibilities such as - when to stock up coffee beans or to schedule cleaning of machine.
Voila! Coffee mission accomplished
Other use cases
This is just a fun project in order to demonstrate how Azure Percept comes in handy and it was extremely easy to build this prototype. There are a variety of other use cases where Azure Percept can come in handy such as finding available meeting rooms, stock detection or defect detection in retail and other scenarios where AI is really powerful.
If you want to get started on developing AI models on Azure Percept:
Other use cases where Azure Percept is used
About me
Gaya works as Data Engineer and has around 9 years of experience in building data engineering and infrastructure solutions for different companies.
When not drinking coffee, she likes to go cycling, hiking and to other outdoor activities