Behind the Scenes: How ADLINK Built Vizi-AI, an Industrial Machine Vision AI Developer Kit

Here at ADLINK, we have been hard at work creating the tools for a more intuitive and simple way to leverage vision as a sensor. Using our experience of building solutions with partners and clients, while leveraging our Edge hardware and simple to use ADLINK Edge Platform, we were keen to make the process quicker and easier for others.

From building proof of concepts with consumer developer kits and additional USB peripherals, to achieve the hardware performance required for vision as a sensor, we quickly realised that this was not practical when trying to run the devices in environments that were experiencing vibration and unusual levels of temperature.

We recognised a need for a developer kit that supported the machine vision AI development process. The vision to create a device was born. A device that was powerful, of a single-board design, that could be used in industrial environments to collect image data and allow a user to send it wherever you need it, to start building and training their AI models.

After training, models are easily returned back to the device to perform the inference, using hardware accelerated capabilities to achieve the decisions to be made quickly for the solution under consideration.

Then, what do you do with the inference data?

How do you enable a simplistic integration to other software as response to the model output data, to deliver on the client’s proof of concept?

How do you easily scale from proof of concept to production?

Keep reading…We will answer those questions a little later.

Across our teams at ADLINK, we have the experience to bring all of the tools together, while also creating some new ones, to enable a more intuitive user experience.

Using our 25 years of experience within our ADLINK Hardware Team, the ADLINK Software & Solutions Team were recommended to leverage the SMARC 2.0 industrial ready board with an Intel® ATOM 3940 CPU, and include the Intel Movidius™ Myriad X VPU for the Vision AI requirements at an appropriate cost to power/inference price point ratio.

Intel Movidius VPUs solve demanding computer vision and edge AI workloads with efficiency. By coupling highly parallel programmable computing with workload-specific hardware acceleration in a unique architecture that minimises data movement, Movidius VPUs achieve a balance of power, efficiency and computer performance. VPU technology enables intelligent cameras, edge servers and AI appliances with deep neural network and computer vision.

The SMARC 2.0 with its 40 pin GPIO 2 USB 3.0, 2 USB 2 ports and 1 Gb Ethernet ports provides the connectivity and options to connect to the network and camera devices required to build out a proof of concept and lead to a solution.

Our ADLINK Edge Solutions Team reflected on the requirements of a number of past projects and solutions they had delivered over the past couple of years.

Our team began to assess what additional tooling would accelerate users’ projects and enable more time on the AI training, where a user’s domain expertise is valuable and requires less time writing code for integration and for common machine vision tasks. Starting with our ADLINK Edge Platform that includes the ADLINK Data River, which moves your data to wherever it needs to be within a network boundary or with additional ADLINK Edge Apps that can cross boundaries to off-site consumers of the data.

The ADLINK Edge Platform is a container-based solution that has been designed to enable users to easily configure and spend little-to-no time on coding. It is an extremely intuitive design and has received strong endorsement from its users.

What tools were needed?

Profile Builder — a tool that is key to the simplicity of the ADLINK Edge Platform as it manages the deployment and operation of the ADLINK Edge solutions.

Stream Viewer — receives video frames and inference results from the Data River and combines them into a single video stream that can be viewed with the Profile Builder, or, a standard 3rd party RTSP Viewer.

Frame Streamer — streams video from an attached camera e.g. USB web cam, or, from an on-device video file.

Intel OpenVINO Inference Engine — reads live video frames from the ADLINK Data River, passing them through the user-defined AI inference model and publishes the results back into the Data River in real-time.

Model Manager — facilitates the AI inference model deployment, by storing models locally for Intel OpenVINO inference engine to load via the ADLINK Data River. Models can be uploaded to the Model Manager via the Profile Builder application.

Training Streamer — facilitates the capturing of images for the purposes of training AI models. Images can be captured to a local file system and/or can be uploaded to an FTP/SFTP server.

Node-RED — the most recent addition to the ADLINK Edge Vision bundle, a commonly used tool in IoT POCs that has been enhanced to provide a simple way to deliver outcomes with your inference data.

We have invested even further in supporting businesses wanting to leverage Edge AI Vision by creating a new community website GOTO50.ai, the website is a place for anyone to learn, share and support each other in this space. With guides,blogs, a community forum and projects demonstrating some of the things that users have made, from automated discos to home security SMS solutions. ADLINK welcomes ISVs to discuss in goto50 about how they have used the ADLINK hardware and software to support their products.

We have accelerated our own proof of concepts, by using Vizi-AI to accelerate machine vision solutions — for example the *ADLINK Smart Pallet. Vizi-AI can help you and your team to accelerate your solutions too, by being very quick to set up and manage, expand and scale out.

• Smart Pallet is an ADLINK Neon-MDX-1000 smart camera, fully loaded with the Smart Pallet software stack that minimises the mistakes of boxes being placed on the wrong pallet and creates a more ergonomically satisfying solution for its users over the cumbersome and more error prone hand scanner options.

What will you teach your Vizi-Ai to see and do? Share it on our Hackster.io platform hub and GOTO50.ai/projects to see what the community is doing.