Want to do Computer Vision as Edge AI?

Chooch AI
4 min readJun 8, 2021

In this all-in-one guide, we’ll show you how to create and set up a new NVIDIA Jetson device and install Chooch AI computer vision models on the device and set up cameras.

Creating a New Edge Device

To create a new edge device [00:16], log into your Chooch.ai dashboard and go to the Devices tab on the bottom left. Click on the orange “Create Device” button in the top right. This will open a modal window with the following settings:

  • The name of your device.
  • The device type (either PC or Jetson).
  • (Optional) The location, description, API endpoint, and MQTT broker.

Once complete, click on the “Create Device” button at the bottom of the window, and click on the newly created device in the Devices table.

Adding a Stream to the Device

To add a stream to the device [01:26], click on the orange “Add Stream” button in the top right. This will open a modal window with the following settings:

  • The stream name.
  • The stream IP, which will likely use RTSP (Real Time Streaming Protocol).

Once complete, click on the “Add Stream” button at the bottom of the window.

Adding a Model to the Stream

To add a model to the stream [01:58], click on the orange “Add Models” button in the top right. This will open a modal window with two steps to complete:

  1. Select the camera to use.
  2. Select the model to use. Clicking on this setting will open a new window for you to select a pretrained or custom Chooch model. Models are classified into directories: for example, the “Public Object” directory contains pretrained Chooch models forobject recognition, while the “My Object” directory contains your own custom trained models. Select your preferred model, and click on the “Add” button at the bottom of the window.

Once complete, click on the “Add Model” button at the bottom of the modal window.

Next, you need to connect your device ID to your Jetson device. Click on the Devices tab, and then click on the “Ubuntu Setup Guide” link on the top left:

  1. Scroll down to step 7 of the guide and copy the code that you will use to download Docker on your Jetson device.
  2. Open the Command Prompt or terminal on your computer and connect to your Jetson device via SSH, e.g.: ssh kasim@192.168.0.114
  3. Paste the three lines of code that you copied from the Ubuntu Setup Guide into your Command Prompt or terminal. The code will automatically execute and prompt you to provide your device ID. You can copy your device ID from the Chooch dashboard under the “Device ID” column. Paste the device ID into the Command Prompt or terminal, and press enter.

Once the Chooch Docker files have finished downloading into Jetson, you can access your Chooch edge control panel at http://localhost:8000. The device ID will already be entered for you automatically in the interface, so simply click on the orange Connect button.

In the Chooch edge control panel, you can view information such as the camera ID, name, IP address, and status; the local IP address; and the current software version. You can also change the device ID by clicking on “Change Device ID” in the top right. If you add or remove a stream or a model in your device, you need to click on the “System Update” button in the control panel to apply your changes.

You can navigate the control panel by clicking on the tabs to the left:

  • The Predictions tab displays the output of your camera and your model. You’ll see two camera feeds: a live feed without detection (on the top) and the model’s most recent detection (on the bottom). In the table to the right, you can see detailed prediction information (e.g. date, score, and class name). Click on the “Json” link to view the prediction in JSON format.
  • The Models tab displays the models that are currently attached to your device.
  • The Reports tab filters your reports by date, event, and camera.

That’s it! Go here to learn more about Edge AI.

--

--

Chooch AI

Chooch AI is an computer vision platform that offers the complete lifecycle from data collection to AI training to AI models on the edge.