Skip to main content

Software Powers Machine Vision

Computer Vision, AI, Machine Vision

Machine Vision systems are becoming pervasive thanks to industrial PCs that incorporate the essential hardware and software building blocks. An uptick in volume is also bringing down the cost of these systems—despite an increase in the number of video channels that can be processed concurrently by a single platform.

These dynamics are helping move real-time video analytics from cloud systems toward the edge. With the help of deep learning inferencing algorithms, vision systems can now identify, analyze, and extract data from live-streamed, recorded, or frame-by-frame video content.

But these systems are just part of an end-to-end video analytics foundation built on flexible, customizable software. With the right tools, organizations can transform their computer vision infrastructure into an evolving, intelligent video analytics engine.

Anatomy of a Surveillance Platform

Machine vision solutions operating at the edge consist of a camera that interfaces with either a gateway or server platform that provides local processing and networking, then transmitting relevant video content or alerts into cloud environments for further analytics or operator review. Figure 1 shows the anatomy of an end-to-end surveillance platform.

A video analytics infrastructure spanning from edge to cloud. (Source: Gigabyte)
Figure 1. A video analytics infrastructure spanning from edge to cloud. (Source: Gigabyte)

Advanced software solutions are required to maximize performance and efficiency across this diverse chain of systems. These solutions must support a range of camera hardware, communications standards, and encode/decode protocols to ensure interoperability, in addition to applying analytics algorithms to video content.


Thanks to OpenVINO integration, one of the greatest benefits of the IVAR platform is improving real-time video processing performance on @Intel CPUs by 1.5x. @insightdottech

Open Up Machine Vision Architectures

A solid foundation for open, interoperable machine vision software is the Intel® OpenVINO Toolkit. OpenVINO is a development platform that accelerates computer vision algorithms built in frameworks like Caffe and TensorFlow across multiple Intel® platforms, including Intel® Movidius vision processing units Intel® FPGAs, and processors with integrated Intel® HD Graphics.

OpenVINO also leverages a common API that abstracts multiplatform programming issues. As a result, machine vision workloads can be easily ported across an OpenVINO-enabled infrastructure, from the cloud/data center to edge servers and gateways down to compatible IP cameras.

Still, OpenVINO is only one part of an end-to-end software solution. The output of OpenVINO-enabled video analytics systems must be presented to operators in an actionable way.

Gorilla Technology has developed the Intelligent Video Analytics Recorder (IVAR), a software-as-a-service (SaaS) video management system (VMS). Spanning from edge to cloud, it is the first IVA software platform certified by the Intel® IoT Solutions Alliance to integrate the OpenVINO Toolkit.

The IVAR software platform ingests images into artificial intelligence (AI) models that can be used to train facial recognition, people counting, vehicle identification, behavioral analysis, and other types of algorithms over time (Figure 2). After the training process is complete, the system streams, records, monitors, and analyzes IP camera footage in real time, then transforms it into actionable insights for operators.

The IVAR platform uses AI to improve the accuracy of machine vision applications. (Source: Gigabyte)
Figure 2. The IVAR platform uses AI to improve the accuracy of machine vision applications. (Source: Gigabyte)

Once captured, the system’s findings can be:

  • Presented as popups or alerts in cloud-based dashboards
  • Exported as log events to third-party systems
  • Recorded to storage devices
  • Used by operators to view specific camera footage for further analysis

Features like fast video search can also be used to quickly analyze a large number of videos and display specific people or the location of objects (Figure 3).

IVAR software can present video analytics in a number of ways. (Source: Gorilla Technology)
Figure 3. IVAR software can present video analytics in a number of ways. (Source: Gorilla Technology)

Thanks to the integration of OpenVINO, one of the greatest benefits of the IVAR platform is that it improves real-time video processing performance on Intel CPUs by 1.5x. Rather than developing complex machine vision systems based on dedicated GPU architectures, developers can use CPU-based designs to significantly reduce cost and time to market.

In fact, they may be able to leverage existing Intel processor infrastructure already deployed somewhere in their video analytics infrastructure.

Endless Integration for Rapid, Real-Time Video Analytics

The IVAR software platform can be easily integrated into existing surveillance systems, enabling numerous cameras at different locations or sites to connect to a central hub.

Take the example of Gigabyte, a supplier of high-performance computing solutions for machine learning applications. It has unveiled a VeMo-powered CCTV system that performs facial recognition, vehicle detection, and behavior tracking across concurrent video streams. IVAR software is the brains of the solution that runs on Gigabyte’s edge server hardware.

OnLogic’s smart NVR design also uses the IVAR platform to perform on-site image processing, taking surveillance applications far beyond passive monitoring. Implementing video analytics on these industrial PCs can enable a variety of new surveillance use cases in smart city, enterprise, and retail environments.

More Than Mere Monitoring

The processing and analysis of high-quality video streams at any point in the machine vision infrastructure can provide significant value in many industries. The flexibility to distribute compute assets across a deployment is key to productivity, efficiency, and cost-effectiveness.

The above demonstrates how machine vision systems can be deployed at much more affordable price points. It also shows how integrating AI natively into a video analytics platform can enhance machine vision applications over time.

With software tools that complement machine vision hardware, vision applications are poised to become more pervasive than ever before. And you can even get them “as-a-service.”

About the Author

Brandon is responsible for Embedded Computing Design’s IoT Design, Automotive Embedded Systems, Security by Design, and Industrial Embedded Systems brands, where he drives content strategy, positioning, and community engagement. He is also Embedded Computing Design’s IoT Insider columnist, and enjoys covering topics that range from development kits and tools to cyber security and technology business models. Brandon received a BA in English Literature from Arizona State University, where he graduated cum laude.

Profile Photo of Brandon Lewis