Selective upload

MICHAEL WHEELER, EDITOR-IN-CHIEF MICHAEL.WHEELER@PHOTONICS.COM

Embedded vision, whether it involves capturing still images or video, is one of the biggest sources of data available today, as more devices become part of the Internet of Things. This has opened up exciting new possibilities for home automation and smart home technology, along with security and personal assistance.

The key enabler is the introduction of new AI-powered processors, allowing for computing at the edge. Because these systems don’t rely on a central data center or consume network bandwidth, vision functions can occur remotely without network access. This overcomes the problem of latency and brings down cost.

In contrast, cloud computing offers vast storage capacity for image data, enabling sophisticated analysis. This includes comparing historical images in order to verify process quality and trace product faults. The downside: Data-driven insight from the cloud doesn’t come free.

Whether a company employs one approach or the other depends on a number of factors related to application requirements.

Selective cloud processing is a third option, as we learn in this issue’s cover story.

Sebastien Dignard, president of iENSO, shares the example of a company that sought to bring an intelligent baby monitor to market, planning to incorporate an inexpensive camera and couple it with proprietary software and public cloud processing. But to check off all the system requirements — such as the ability to recognize changes in a baby’s vital signs and quick responsiveness, while achieving a low price point for the consumer — neither sending all the data to the cloud nor moving all the computing power to the edge was feasible.

Instead, an adaptive system was built that could offload data and processing between the edge and the cloud on demand — an approach we’re likely to see in other incarnations. For details on how it was done and other potential use cases, click here.

We hope you enjoy the issue!
>