Machine Vision & Learning


According to the Automated Imaging Association (AIA), machine vision (MV) encompasses all industrial and non-industrial applications in which a combination of hardware and software provides operational guidance to devices in the execution of their functions based on the capture and processing of images. MV systems rely on digital sensors protected inside industrial cameras with specialized optics to acquire images, so that computer hardware and software can process, analyze, and measure various characteristics for decision making.

What Isn’t Working

Industrial MV uses many of the same algorithms and approaches as academic, governmental, and military applications of MV, however, the constraints are different. Industrial MV systems demand greater robustness, reliability, and stability compared to an academic MV system and typically cost much less than those used in governmental or military applications. An ideal industrial MV system must support low cost, low accuracy, high bandwidth, high robustness, high reliability, and high mechanical and temperature stability. As edge computing becomes more prevalent, IIoT enterprises utilizing MV would likely require a larger number of network endpoints (e.g., MV sensors) outside of DMZs, VPNs, firewalls and other technologies built to secure legacy business applications in the cloud. This widens the enterprise’s attack surface. Since the legacy data network and security architectures are not equipped to handle the MV security needs at the edge of the network, enterprises need to address this gap with alternative technologies..

The Way Going Forward

As Gartner recommends, today’s digital enterprise needs a worldwide fabric / mesh of network and network security capabilities to connect entities to the networked capabilities they need access to, when and where they need them.

Enterprises must also be able to leverage context-aware routing between the local sites where MV sensors (or cameras) are deployed and a hybrid cloud environment where the ML applications reside. Since today’s ML models rely on continuous learning in real time, such a network should ideally enable dispersed devices to be programmable over the air in real time. Basic firewall, identity, and access management as well as audit, access & accounting policies should be built in.

Lastly, such a MV/ML friendly security architecture must be able to offer a user-friendly single-pane-of-glass interface to facilities operators for seamless manageability and usability.

Key Takeaway

Instead of relying on legacy technologies that simply are unable to scale for the emerging world of machine vision at the edge, an identity-centric, edge-computing friendly, overlay security solution that is agnostic to the last mile access connectivity networks with foundational support of ML and MV provides a unified and globally connected security network infrastructure. This dramatically simplifies the security needs of the dispersed MV networks aiming to drive business efficiencies with the disruptive power of the ML.

Didn't find what you were looking for?

Contact us and we’ll get back to you as soon as possible.

Contact Us