It’s strange to think that, less than a year ago at the time of this writing, the vast majority of people around the globe had never even heard the term social distancing. But then COVID-19 appeared on the scene, quickly developing into a worldwide pandemic, and social distancing became the norm.

Companies and organizations want to allow people to return to physically shared spaces so long as there aren’t too many people in the space and those people follow social distancing protocols. What is required is some way to measure occupancy and monitor social distancing in real time.

The solution is to use computer vision (CV) combined with artificial intelligence (AI) to detect people, determine their number, and measure the distance between them.

WE WANT MORE!

It’s a general truth that everybody wants more; for example, designers of electronic systems want more capacity and performance, while users want more features and functions.

As it happens, AI systems are designed to provide more. They pass inferences back into training models to add more intelligence, which delivers more accuracy and more efficiency and even the potential to interpret more unique types of data.

The people who design and use AI systems also want more. For example, the clients of systems integrators building CV-based social distance monitoring systems for public spaces realized that this same infrastructure could be used to perform people counting, occupancy monitoring, and other similar AI-based functions.

Unfortunately, by the time it occurs to everyone that such capabilities should be added to a system design, hardware has been selected and software is being optimized, so it is usually too late. That is, unless they are using an AI edge infrastructure platform with the ability to incrementally add AI models – perhaps even run multiple models concurrently – at the edge.

SCALING VISION INFERENCE AT THE EDGE WITH VIZI-AI

Of course, such edge infrastructure platforms aren’t easy to find. While the value these systems can deliver is undeniable, the remote, distributed nature of edge environments means that the systems deployed there are typically constrained in terms of power consumption, size, and cost. As a result, many do not contain the compute and memory performance required to run sophisticated AI models, much less run multiple models concurrently.

In addition, the diversity of application requirements at the edge makes scaling systems en masse a problem from both a hardware and software perspective (Figure 1).

(Figure 1. Several challenges must be overcome to scale edge computing platforms to the demands of multi-algorithm AI inferencing (Source: ResearchGate))

In light of these obstacles, one potential is the Vizi-AI Development Starter Kit (Figure 2). Vizi-AI combines plug-and-play hardware and software, enabling a faster, easier, and scalable starting point for machine vision AI deployments at the edge.

(Figure 2. The Vizi-AI Development Starter Kit. (Source: ADLINK Technology, Inc.))

The Vizi-AI Development Starter Kit features a SMARC that is powered by an Intel® Atom processor containing a CPU and GPU, as well as an Intel Movidius™ Myriad™ X VPU. The Movidius Myriad X VPU features a dedicated hardware accelerator for deep neural network inferencing called the Neural Compute Engine, which means that more demanding or priority inferencing workloads can run on the VPU while other inferencing models can run on the Atom processor’s GPU or CPU. In turn, this means that concurrent inferences can be performed on a single video or image stream.

The SMARC computer module at the heart of the kit is a complete embedded computer system implemented on a small PCB that employs an industry-standard interface specification. The SMARC module supports I/O specifically designed for applications like machine/computer vision, simplifying the integration of Luxonis DepthAI cameras, Basler cameras, or other GigE cameras, some of which contain their own processing elements that can be used to provide even more layered AI intelligence.

One example of real-world social distancing monitoring using the Vizi-AI was recently published by Toby McClean, who demonstrated the platform in use with the aforementioned Luxonis DepthAI: USB3C camera. The entire system can be built in just a few hours, thanks in large part to the plug-and-play software made available with the Vizi-AI kit, including:

• The Intel Distribution of the OpenVINO™ toolkit, which optimizes deep learning workloads across Intel architecture -- including accelerators -- and streamlines deployments from the edge to the cloud.
• ADLINK Edge™ software, which provides enhanced functionality of OpenVINO through the ADLINK Data River™ that enables data to flow freely and securely.
• ADLINK Edge™ Profile builder, which provides a simple, intuitive user experience to manage devices and applications.
•  ADLINK Edge™ Model Manager, which enables users to add their own models to a pre-loaded selection for easy deployment.

Together, these and other components make up the ADLINK Edge software development kit (SDK). The Edge SDK is a set of libraries and tools that allow developers to quickly and easily build IoT solutions using simple APIs for creating and maintaining “things” on a data sharing framework called the ADLINK Data River. This Data River simplifies complex data network programming and enables data to flow freely, but securely, northbound or southbound to any cloud analytic platform database, and even east to west between devices, databases, and the cloud. All of this activity is orchestrated using ADLINK’s version of Node-RED, which is an open source UI-based programming tool for easily wiring together flows of data.

Also available are a set of pre-trained models that users can use for learning and demonstration purposes or for developing their own deep learning software. These models can be accessed via the OpenVINO Model Zoo on Github.

In the tutorial mentioned above, McClean implements all of the logic necessary for people counting, distancing monitoring, and a visualization dashboard in the easy-to-use Node-RED language (Figure 3). Using a DepthAI class to pull data from the camera, he is able to acquire, manage, and transport data onto the Data River and into nodes that perform the various functions listed previously.

(Figure 3. With ADLINK Node-RED, developers are able to generate outcomes from vision-based inferences using an easy and intuitive programming model.)

SELF-SUFFICIENT BUT LOVES TO MAKE FRIENDS

The Vizi-AI Development Starter Kit can be deployed as a self-contained industrial machine vision AI at the edge, which makes it ideal for applications that demand low latency, data privacy and security, and low networking costs. However, the system can also be easily integrated with AWS services for deployments that require additional processing and analysis in the cloud.

It’s also important to note that Vizi-AI is designed from the ground up for industrial use cases, being based on the industry-standard SMARC COM architecture. The ability to replace one SMARC COM with another while keeping the same carrier board allows scalability, fast time-to-market, and upgradability, while maintaining low costs, low power, and small physical size. Furthermore, should it be deemed that special hardware optimizations are required, Vizi-AI’s compatibility and support for Intel’s MRAA library enables maximum software portability, even in the event that components change.

Of particular interest to developers is the fact that Vizi-AI is supported by its own community at https://goto50.ai/ where users can find support, pre-built scenarios, projects, and other useful resources.

Like Vizi-AI, we all love to make friends, but the latest surge in the COVID-19 pandemic means we need to be careful. Sometimes we forget ourselves, so using an ADLINK Technology’s Vizi-AI edge-based AI platform to measure occupancy and monitor social distancing can literally save lives.

ABOUT ADLINK  

ADLINK is a global leader in edge computing, a Premier Member of the Intel® Internet of Things Solutions Alliance, and a contributor to standards initiatives such as Eclipse, OCP, OMG and ROS2 TSC. ADLINK is ISO-9001, ISO-14001, ISO-13485 and TL9000 certified and is publicly traded on TAIEX (Stock Code: 6166). 

ADLINK’s mission is to affect positive change in society and industry by connecting people, places and things with AI. Their offerings include robust boards, real-time data acquisition solutions and application enablement for AIoT. ADLINK serve vertical markets including manufacturing, communications, healthcare, military, energy, infotainment and transportation.