Deploying Computer Vision at Edge Scale — Part 1
Sunlight Founder & CEO, Julian Chesterfield shares his insights with Edge Computing Expo, North America.
Why is Computer Vision at the edge so important? What is needed to roll it out economically at scale? As an edge infrastructure company working with Computer Vision companies with distributed edge deployments, we have put together this two-part blog series to answer these questions for you.
The rise of Computer Vision
We’ve all seen the huge growth in Computer Vision use cases over the last couple of years. Examples include self-driving cars, quality control systems in manufacturing, number plate recognition in drive thrus, interpretation of medical lab results, billing for low emission zones in cities, and moving freight around warehouses. These use cases help enterprises to increase efficiency, reduce downtime, and improve yields, whilst lowering overall costs.
In an age where many industries are facing labour shortages, the ability for enterprises to automate without impacting customer satisfaction is also a prime driver. What’s making all these exciting use cases possible, is the ever-increasing range of software products that use artificial intelligence and machine learning to interpret the video streams and even take actions autonomously. In fact, the analyst firm IDC estimates a 57% compound annual growth rate for the Computer Vision market up to 2025.
Use of Computer Vision will impact nearly every industry and provides the missing piece between connecting the physical and virtual worlds in real-time.
The need for edge processing
Up until recently, a lot of the computational heavy lifting for these applications was done ‘in the cloud’. The hyperscalers, such as AWS, Microsoft Azure and Google Cloud Platform, can make almost unlimited computing resources available, and even have Artificial intelligence tool sets which can simplify development.
However, as Computer Vision applications get more sophisticated and rely upon higher fidelity data, pushing that data backwards and forwards to the cloud becomes a problem — especially if you’re in a remote location like an oil platform, or on a farm, or where there is intermittent connectivity like a city bus, or there is a requirement for specific security needs — like an air-gapped production line system.
Not only can bandwidth be insufficient, but the network communication costs rack up, and round-trip times to the cloud mean that implementing real-time response to the data is impossible.
One of the biggest issues we see is ensuring each location can operate autonomously, in other words making sure Quality Control systems or Point-of-Sale systems don’t stop working just because it can’t reach the cloud.
Now multiply that problem by thousands of locations and you’ll see why pushing compute power to where the video data is generated is so important. That is why the foundation for Computer Vision at the edge needs a massively scalable, yet tiny, secure and efficient infrastructure that can be centrally managed.
An infrastructure for Computer Vision at edge scale
First let’s cover off one of the most important non-technical aspects of edge infrastructure — cost. According to the analyst firm ESG, the cost of deploying infrastructure to edge locations is the top concern among enterprises. That’s not surprising when you consider that 96% of enterprises have two or more applications in each edge location, with 37% of enterprises deploying more than five. We’re already seeing ‘edge sprawl’ — where each application has its own discrete physical hardware. It’s early days for edge applications, so that number will only grow. This is expensive — both in terms of hardware costs and management costs. It’s far better to deploy a general-purpose edge computing infrastructure that can support all your edge applications in a single physical stack. We will be looking at this in part two of the series and sharing the requirements you need to consider for a cost-effective Computer Vision infrastructure that scales with you. Read part 2.
