search

IBM / node-red-tensorflowjs

Node-RED node with a TensorFlow.js Object Detection model
Apache License 2.0
109 stars 60 forks source link

readme

Build Status

Developing a Machine Learning IoT App with Node-RED and TensorFlow.js

In most cases, enabling your IoT device with AI capabilities involves sending the data from the device to a server. The machine learning calculations would happen on the server. Then the results sent back to the device for appropriate action.

When data security or network connectivity is a concern this is not an ideal or feasible approach.

With this code pattern, you will learn how to build and deploy machine learning apps that can run offline and directly on the device (in this case a Raspberry Pi). Using Node-RED with TensorFlow.js you can incorporate machine learning into your device is an easy, low-code way.

Node-RED is an open source visual programming tool that offers a browser-based flow editor for wiring together devices, APIs, and services. Built on Node.js, you can extend its features by creating your own nodes or taking advantage of the JavaScript and NPM ecosystem.

TensorFlow.js is an open source JavaScript library to build, train, and run machine learning models in JavaScript environments such as the browser and Node.js.

Combining Node-RED with TensorFlow.js developers and IoT enthusiasts can more easily add machine learning functionality onto their devices.

When you have completed this code pattern, you will understand how to:

architecture

Flow

  1. The user owns or downloads a machine learning model in TensorFlow.js format.
  2. The user creates a Node-RED node for the TensorFlow.js model and wires the TensorFlow.js node in a Node-RED application.
  3. The user can deploy the Node-RED application locally.
  4. The user can access the Node-RED application from a browser and can trigger inferencing on images captured from a webcam.
  5. Alternatively, the user can deploy the Node-RED application to a Raspberry Pi.
  6. The device runs the Node-RED application and performs inferencing on images from a camera.
  7. The device can output to a connected speaker or take some other action depending on the inference results.

Included Components

Featured Technologies

Watch the Video

Prerequisites

Steps

Follow these steps to setup and run this code pattern. The steps are described in detail below.

  1. Clone the repo
  2. Install dependencies
  3. Import the Node-RED flow
  4. Deploy the Node-RED flow

1. Clone the repo

First let's get the code. From the terminal of the system you plan on running Node-RED from, do the following:

  1. Clone the node-red-tensorflowjs repo:

    $ git clone https://github.com/IBM/node-red-tensorflowjs

  2. Move into the directory of the cloned repo:

    $ cd node-red-tensorflowjs

Note: For Raspberry Pi users, details on accessing the command line can be found in the remote access documentation if not connecting with a screen and keyboard.

2. Install dependencies

You can install the necessary dependencies by running:

$ npm install

This will install Node-RED along with any necessary custom node packages for running the browser flow in the local node_modules folder, and you can move on to starting Node-RED.

Alternatively, if you already have Node-RED installed on your system, you can just install the dependencies from your Node-RED user directory (~/.node-red). Run the following block of code, being sure to change the <full path> placeholder to the path of the cloned repo:

cd ~/.node-red
npm install <full path>/node-red-contrib-tfjs-object-detection
npm install node-red-contrib-browser-utils node-red-contrib-play-audio node-red-contrib-image-output

Be sure to restart Node-RED if it was already running when installing this way.

Note: If you are using a Raspberry Pi, instructions for installing Node-RED can be found here. However, if you are using the Raspbian operating system for the Raspberry Pi, Node-RED comes pre-installed, so you can just install dependencies from the ~/.node-red directory.

Start Node-RED

Node-RED can be started from a terminal by running this command from within the directory of the cloned repository:

$ npm start

Alternatively, if you have Node-RED installed globally with dependencies installed under ~./node-red, you can start Node-RED from any directory:

$ node-red

Stop Node-RED

You can stop Node-RED by closing the terminal window or using Ctrl-C in the terminal.

Node-RED editor

The Node-RED editor can be accessed from http://localhost:1880.

However, if Node-RED is on the Raspberry Pi, you can connect to it via http://<Raspberry Pi IP>:1880.

3. Import the Node-RED flow

Once installed the node can be added and used in the flow of your Node-RED application. To import the flows available in this repo:

  1. Make sure Node-RED is running
  2. Open a browser and go to your Node-RED Editor
  3. Click on the Node-RED Menu
  4. Click on Import
  5. Select the Clipboard tab
  6. Click on select a file to import
  7. Browse to and select one of the flow files in the cloned repo
    • If trying things out locally on your browser, then use the browser-flow.json.
    • If using a Raspberry Pi with peripherals, then use the raspberrypi-flows.json.
  8. Select Import to new flow
  9. Click Import

4. Deploy the Node-RED flow

The Node-RED flow can be deployed in multiple ways. Follow the option that best fits your use case:

Running on a Raspberry Pi

The Raspberry Pi flows use hardware peripherals and Raspberry Pi specific nodes. This assumes you imported the raspberrypi-flows.json file.

Pre-requisites

The following hardware components are needed to fully run this flow:

Additionally, a few custom nodes are needed and can be added through the Palette Manager:

Deploy and run on a Raspberry Pi

The imported flows file contains two flows:

Make sure all your hardware is connected, then:

  1. Double click on Play Audio File exec node and change the path in the append section to the path of a .wav file of your choosing. Click Done when finished.
  2. Click the Deploy button.
  3. Trigger the camera.
    a. Can manually trigger the snapshot by clicking the Take Photo inject node.
    b. If using the motion sensor flow, motion near the sensor will trigger the camera.

Note: Feel free to change the detected object by editing the code in the isObjectDetected node.

Running on a laptop or workstation

From the Node-RED editor, do the following:

  1. Click the Deploy button.
  2. Upload or capture an image.
    a. Click the file inject node and browse to an image.
    b. Click the camera node and allow the browser to access the webcam.

The image will be processed by the tfjs object detection node and the output will be displayed in the Debug panel. If the browser supports the Web Audio API, the objects detected will be spoken.

Output

Raspberry Pi flow

raspberry pi flow

Browser flow

browser flow

Links

License

This code pattern is licensed under the Apache License, Version 2. Separate third-party code objects invoked within this code pattern are licensed by their respective providers pursuant to their own separate licenses. Contributions are subject to the Developer Certificate of Origin, Version 1.1 and the Apache License, Version 2.

Apache License FAQ