Embedded Vision Drives Innovation Across Industries

Embedded vision technology is poised to permeate almost every facet of daily life. A recent panel at Embedded World 2021 explored its current deployment, AI integration, and edge‑cloud dynamics.

Advances in sensors, processors, and software are driving embedded vision into agriculture, manufacturing, autonomous vehicles, and professional sports. The COVID‑19 pandemic further accelerated its deployment, with vision systems now used for public surveillance, health checks, and safety inspections.

AI‑Enabled Embedded Vision

Artificial intelligence is reshaping image processing as developers increasingly adopt deep learning and neural networks to enhance object detection and classification. AI unlocks new possibilities, yet panelists agreed it must become easier to use.

“AI offers significant benefits to customers, but it can feel clumsy,” said Olaf Munkelt, managing director of MVTec Software. “We need to simplify the entire AI workflow—from data labeling and inspection to management and deployment—so that embedded vision solutions deliver tangible value quickly.” Munkelt called for an integrated approach that streamlines each step, including semantic segmentation, classification, and anomaly detection.

Fredrik Nilsson, head of the Machine Vision unit at Sick, echoed this sentiment. He noted that deep learning solves tasks that rule‑based image processing struggles with, but the two approaches will coexist for the foreseeable future. “Hybrid solutions—using deep learning for segmentation and traditional algorithms for measurement—will become standard,” Nilsson said.

Munkelt also highlighted the race in AI accelerator hardware. Start‑ups are producing chips that can outperform established GPU vendors by 10–20×. Speed will become critical as image volumes grow, and the vision community is actively exploring these accelerators for maximum performance.

What Happens on the Edge? What Happens in the Cloud?

Amazon Web Services (AWS) outlined a dual strategy for embedded vision. Austin Ashe, head of strategic OEM partnerships for IoT, said AWS aims to lower the entry barrier for newcomers and to extend value beyond initial pilots. “Approximately 75 % of businesses plan to move from pilot to full operation within the next two to five years,” Ashe said, highlighting the need for orchestration between edge and cloud.

Edge devices handle latency, bandwidth, cost, and security. AWS can monitor devices—whether a single unit or a fleet—providing real‑time alerts and enabling over‑the‑air updates. This model allows companies to train a model in the cloud and then deploy it across their machines seamlessly.

For organisations lacking data‑science talent, Ashe explained that uploading just ten to twelve anomaly images to the cloud can instantly produce a custom detection model, which can then be iterated and pushed to the edge.

During the conference, Basler and AWS demonstrated how their partnership bridges the gap with AWS Panorama—a machine‑learning appliance and SDK that empowers real‑time visual decision‑making—and Amazon Lookout for Vision, a service that identifies defects and anomalies in images.

As applications demand lower latency, Ashe noted that edge will remain paramount, but emerging 5G networks will tighten the integration between cloud and edge, opening new use‑case possibilities.

Complexity, Size, and Cost

Arndt Dake, CMO of Basler, warned that the multi‑processor landscape—CPU, GPU, AI accelerator, ISP—adds complexity. “Customers must map software across four resources instead of one,” he said, stressing that usability and demonstrable benefits drive adoption.

On device size, Dake said the industry will continue to shrink, with smartphones as the benchmark for compact, high‑performance vision systems.

Cost remains a barrier. Munkelt noted that many applications are not yet justified due to high prices; reducing costs will unlock new possibilities. Nilsson added that easier use, lower prices, and smaller devices will make embedded vision accessible to smaller companies that have not yet embraced the technology.

>> This article was originally published on our sister site, EE Times Europe.