This article provides you with a checklist of prerequisites and device configuration steps about how to set up your environment before you begin to create your first end-to-end Edge (artificial intelligence) AI solution with Azure DeepStream Accelerator. There are five key steps you must complete:
- Step 1. Software and model prerequisites
- Step 2. Install Azure Services
- Step 3. Set up the Edge device
- Step 4: Container deployment: Create a deployment manifest file
- Step 5: Deploy to your Edge device
Complete the following software and AI-model items to configure your environment:
-
Clone the repository to your local environment.
-
Install the following:
-
An active Microsoft Azure subscription with required permissions.
-
A RTSP stream player (VideoLAN VLC media player or similar).
-
A trained computer vision model. Depending on your model selection, you may also be required to supply a parser. For information about which parsers are available in this repository or from other sources, visit Azure DeepStream Accelerator overview.
You must deploy all of the following Azure Services:
- Azure Container Registry, where you will store your custom Docker containers.
- Azure IoT Hub, which will manage the IoT Edge solution.
- Azure Storage, which will be used to store your videos.
- See here for deployment instructions
- Also create a container instance inside the Azure Storage Account (see here for instructions). This container will store your video snippets.
- Don't confuse this Azure Storage Container with a Docker Container or with the Azure Container Registry. The latter two refer to Docker Containers, while the Azure Storage Container is also called a blob storage container - it is for storing arbitrary data (but not Docker Containers!)
Install the CLI tool. For information, visit the CLI Tool documentation.
-
In the Azure portal, go to your Azure Storage Account.
-
From the left navigation, select Access Keys.
-
Find key1 and select Show.
-
Copy the value of key1 and save it. You'll need it shortly.
-
Go to your IoT Hub in the Azure portal.
-
On the left navigation, under Hub Settings, locate File Upload.
-
Click Select Azure Storage Container.
-
Then select your Blob Storage Account.
-
Save.
Azure DeepStream Accelerator is available on any device with sufficient storage space and an NVIDIA GPU, as long as it supports NVIDIA's container runtime. However, most of our testing is done on Jetson AGX Orin. Please be advised that we cannot handle issues opened in regards to untested setups.
If you would like to try the experience with a discrete NVIDIA GPU, it should work, but please be aware that we do not test this configuration as heavily as the Orin setup, and we may not be able to help with issues related to this setup.
- See here for an overview on setting up Azure IoT Edge.
- See here for a single Linux device setup.
The NVIDIA Jetson AGX Orin Developer Kit allows you to develop full-featured AI applications for products based on Jetson devices.
- For information on how to provision your NVIDIA Jetson device to Azure and ensure that you have all the needed applications and libraries in your device, visit Azure IoT Edge Configuration tool.
- For device-specific documentation, visit: Getting Started with Jetson AGX Orin Developer Kit.
After you have deployed the basic Azure IoT environment, your next step is to deploy the Azure DeepStream Accelerator solution.
To deploy your Azure DeepStream Accelerator solution, you need the following five modules (Docker containers). These Docker containers work together to run the computer vision (CV) model, detect relevant business events, and upload relevant video to the cloud for future use.
- AI Pipeline module: Contains the DeepStream compatible CV model and communicates model inferences to the Business Logic module.
- Controller Module: Enables management of the entire solution.
- Business Logic module: Contains event detection logic, based on user defined Regions of Interest (ROIs). This is usually where you put your own business logic code (if applicable). It will determine if a pre-defined event occurs during a video segment and communicate this status to the Video Uploader.
- RTSP Simulation module: Converts a stored video file into an RTSP stream to facilitate development. This is useful for debugging and development, as it allows you to use a video file instead of having to set up a camera for development. This container is optional in a deployment.
- Video Uploader module: Processes video segments of pre-defined length. Uploads only those segments containing pre-defined events to blob storage, based on input from the Business Logic module.
These modules are deployed using a deployment manifest JSON file. The manifest file is created by populating the manifest template with environmental variables, contained in an .env file. The file may contain confidential information. Be careful that you don’t inadvertently share this information.
Here is an example of an .env file. Note that since it is an example, it may not accurately reflect the most recent version of the file.
CONTAINER_REGISTRY_NAME=your-container-registry-url
CONTAINER_REGISTRY_PASSWORD=your-container-registry-password
CONTAINER_REGISTRY_USERNAME=your-container-registry-username
BLOB_STORAGE_ACCOUNT_KEY=""
APPLICATIONINSIGHTS_CONNECTION_STRING=""
IOTHUB_DEVICE_CONNECTION_STRING=your-device-connection-string
DS_AI_PIPELINE_IMAGE_URI=mcr.microsoft.com/azureedgedevices/azdaaipipeline:d923853-amd64
BUSINESS_LOGIC_IMAGE_URI=mcr.microsoft.com/azureedgedevices/azdabusinesslogicmodule:d923853-amd64
Note
If you’re using an x86 device, use this manifest template and modify the values in the template.env file.
If you’re using an ARM device, use this manifest template and modify the values in the template.env file. Further,
if you'd like to use a USB camera, you will find this manifest template to be useful.
Note
You should copy the template.env file and change its name to ".env" - not "template.env". The IoT extensions for VS Code assume that you have a file literally named ".env" in your deployment manifest template's directory.
To set up your environment, you must make two changes to the .env file:
- Replace the value of
IOTHUB_DEVICE_CONNECTION_STRINGwith the value from the Azure portal. - Replace the value of
BLOB_STORAGE_ACCOUNT_KEYwith the value you saved in the Get your Blob Storage Account Key section, earlier in this article.
Leave the other environment variables in the file with a blank or null value for now. We’ll fill them in in later tutorials.
- In VS Code, right-click the deployment manifest template and select Generate IoT Edge Deployment Manifest.
This generates a JSON manifest file, which will be saved in a configuration folder in either ds-ai-pipeline/arm or ds-ai-pipeline/x86_64.
Now that you have a JSON manifest file, the next step is to create a deployment to your device.
-
Your JSON manifest file is stored in a
configfolder in eitherds-ai-pipeline/armords-ai-pipeline/x86_64. You can set this deployment to your device. -
In VS Code, right click on the generated deployment manifest JSON file you created (NOT the template), select Create deployment for single device, and then select your IoT Edge device.
Now that you have completed setting-up your environment, you are ready to start creating your first end-to-end Edge AI solution leveraging Azure DeepStream Accelerator. We recommend the following tutorial to get started: