Schedule Attendees Get Involved FAQ

Hosted by

Louis Moreau

Senior DevRel Engineer - Edge Impulse

Learn more

Build a people counter with an MCU-based camera device

Where

Workshops Area 1

When

09/23/2022 1:00pm

Friday 23 September 2022
04:00 am - 05:00 am PT

Add to Calendar 09/23/2022 1:00pm 09/23/2022 2:00pm Europe/Paris Build a people counter with an MCU-based camera device

Counting people is one very common use case in IoT applications. Companies use the results techniques to make sure a store or an office is not too crowded, how many people are walking in front of an advertising panel, or for intrusion detection.

Several low-cost sensors can be used to detect the presence of a person. But when it comes to detecting the number of people, it is usually much more complex and, usually requires a lot of computing resources or high-end sensors.

In this workshop, we will see how to build, train and deploy a custom tinyML application able to count people and send the inference results through TheThingsNetwork using LoRaWAN connectivity. We will be using an MCU-based device with a low-resolution camera sensor to run the custom object detection model.

By using a tinyML model with LoRaWAN connectivity, the frames are processed locally and only the inference results are forwarded to the cloud, respecting the people’s privacy.

Summary of the content of the workshop:

Introduction

Build your machine learning pipeline

  • Collect data
  • Create a custom ML pipeline
  • Preprocess your data
  • Train your machine learning model
  • Validate your model

Run the inference on an Arduino Portenta H7 + LoRaWAN vision shield

Counting people is one very common use case in IoT applications. Companies use the results techniques to make sure a store or an office is not too crowded, how many people are walking in front of an advertising panel, or for intrusion detection.

Several low-cost sensors can be used to detect the presence of a person. But when it comes to detecting the number of people, it is usually much more complex and, usually requires a lot of computing resources or high-end sensors.

In this workshop, we will see how to build, train and deploy a custom tinyML application able to count people and send the inference results through TheThingsNetwork using LoRaWAN connectivity. We will be using an MCU-based device with a low-resolution camera sensor to run the custom object detection model.

By using a tinyML model with LoRaWAN connectivity, the frames are processed locally and only the inference results are forwarded to the cloud, respecting the people’s privacy.

Summary of the content of the workshop:

Introduction

Build your machine learning pipeline

  • Collect data
  • Create a custom ML pipeline
  • Preprocess your data
  • Train your machine learning model
  • Validate your model

Run the inference on an Arduino Portenta H7 + LoRaWAN vision shield