Jetson Project of the Month: Automate Your Household Duties with NVIDIA Jetson and DeepStack

The Jetson Project of the Month simplifies home automation projects using a combination of DeepStack and Home Assistant along with the NVIDIA Jetson.

The holidays should be a time for relaxation, and with NVIDIA Jetson technology, you can do just that. The latest Jetson Project of the Month comes from a developer who has created ways to simplify home automation projects, using a combination of DeepStack, Home Assistant, and NVIDIA Jetson. 

Robin Cole, a senior data scientist at Satellite Vu with a background in physics, developed this project. He has a passion for problem solving using data and Python. Cole’s main activities on GitHub involve the practical application of machine learning. 

As he notes on his GitHub profile: “I have a number of personal projects around training and deploying neural networks on edge devices such as the Raspberry Pi and Jetson Nano.”

In his spare time, he is actively involved in two community projects: Home Assistant and DeepStack AI, and is exploring ways to combine the two technologies. Cole’s Jetson project integrates DeepStack object and face detection and recognition services into the popular home automation platform Home Assistant.

DeepStack runs in a Docker container and exposes various computer vision models through a REST API. DeepStack object detection can identify 80 different kinds of objects, including people, vehicles, and animals. The current list does not include sleighs or reindeer, but you can use a custom object detection model. DeepStack is free to use and it is fully open source. You will need a machine with 8GB RAM or an NVIDIA Jetson to run DeepStack.

Home Assistant is an open-source and extensible platform for home automation that can run on local embedded devices or servers while maintaining local control and privacy. It can help automate tasks such as home energy management, intelligently turning on or off the lights, or sending alerts based on entrances being open. Home Assistant can be integrated with more than 1,000 different devices.

This project uses trained models in the DeepStack REST API to send responses to Home Assistant, which can then perform tasks around the home based on this and other input. DeepStack can run on a Jetson device and it achieves good inference times on its YOLOv5-based trained models, which are run with PyTorch. The DeepStack and Home Assistant mediate their communication through a Go server and a Redis layer.

“I hope this project inspires readers to think about novel applications in and around the home and business that could benefit from AI, and enables rapid prototyping and experimentation,” Cole wrote in a forum post describing his work.

These home automation processes could be applied to tasks like monitoring the Christmas tree lights, issuing alerts when special packages are delivered to the front door, or keeping an eye on the chimney to see if a big guy in a red suit tries to sneak inside the house late at night and eat all of your cookies.

Think of it as a spin on the old holiday classic: You can keep a lookout when you’re sleeping; there’s no need to be awake. Your home automation setup can do the work, which is certainly good for goodness’ sake

The folks at Everything Smart Home put together demos of each part of this project:

Person and face recognition

Object detection

By combining DeepStack with Home Assistant, Cole has shown developers how to create their own home security or automation systems and the ability to integrate Jetson into larger projects. For those who want to dig deeper, this project opens further possibilities to integrate other NVIDIA technologies such as DeepStream, Metropolis, or even NVIDIA Isaac with Home Assistant.

As Cole points out in his forum post about this project, the developer community is all in and finding smart ways to use this DeepStack-Home Assistant combo, including:

  • Monitoring activity in a brick factory in Latin America.
  • Watching for intruding snakes in Thailand.
  • Monitoring parcel deliveries.
  • Checking that a motorcycle is locked.
  • Checking when a chicken lays an egg.
  • Greeting people when they return home and playing a theme tune.
  • Counting visitor numbers at a shop.
  • Checking when a parking spot has become available.

The possibilities are only limited by your imagination. We still think a setup to keep tabs on Santa could be very handy. Share your ideas for how to combine DeepStack and Home Assistant with NVIDIA Jetson in the developer forums. 

Cole has also been developing a custom model to detect fires using camera feeds.

Details about Cole’s projects featured in this post are available on GitHub: object detection and face recognition.

You can see a more detailed demo of Cole’s integration of DeepStack with Home Assistant in this presentation he made at IceVision earlier this year.

Learn more about using DeepStack with NVIDIA Jetson.

Source:: NVIDIA