Skip to main content Skip to footer

Article

Vision Inspection for Industrial IoT. Powered by Edge Analytics.

Using video cameras as sensors in industrial IoT applications is becoming more and more popular. The reason is of course that it opens up for many interesting applications, such as:

  • Vision Inspection for Yield optimization - Identify scrap material and/or products as early as possible in the process
  • Count and measure - Count products, objects, people and measure position, alignment, color and other attributes
  • Intrusion detection - Trigger an alarm or stop a machine when people or objects get too close or enter where they shouldn’t be.

The advantages of cameras as sensors:
Smart, Fast and Low Cost

One of the main advantages of using cameras as sensors is that they are non-intrusive. They operate from a distance and don’t require physical contact with the objects or features you want to monitor. In addition, a vision-based monitoring system can be added independently of any existing monitoring/IoT setup and you don’t need to involve the OT systems. Vision systems are therefore often fast to deploy and to get a return on the investment.

They can also be used to “upgrade” legacy machines and assets. Analogue lamps, switches and relays that needs to be periodically controlled can be monitored by cameras with smart workflows and actions when there are changes.

The downside. And how Edge Analytics can be the solution

The downside is that video cameras produce a lot of data, and if actions are needed based on the video feed, latency will also be an issue. This means that processing in the cloud is in most cases not an option.

Fortunately, Crosser’s Edge Streaming Analytics solution is a perfect environment for implementing vision algorithms close to the cameras and to build smart workflows to take action with notifications, triggers to machines or integrations to enterprise systems such as ERP systems.

Crosser Vision Inspection Use Case

What you need for the solution

Everything you need is easily available today, including:

  1. Off-the-shelf cameras
  2. Off-the-shelf hardware platform (Raspberry Pi and up)
  3. Self-service Edge Analytics software
  4. Vision algorithms

With the Crosser Flow Studio graphical design tool you can easily implement your vision algorithms and deliver the results to on-premise systems or cloud services using modules from the Crosser library.

As an example, you can use the Python module to run your algorithm based on the popular OpenCV library. The OpenCV (Open Source Computer Vision Library) is an open source computer vision and machine learning software library that has more than 2500 optimized algorithms, which includes a comprehensive set of both classic and state-of-the-art computer vision and machine learning algorithms.

If you have trained your own machine learning model in TensorFlow or some other ML framework you can execute the model on the Crosser node close to the camera.

The Crosser Edge Director helps you deploy your Flows and Algorithms to any location in the Edge or Cloud with ease. If you have deployed the same logic on multiple locations the Edge Director lets you mass-deploy and update versions on all nodes in one single operation.

Recommended reading: ML in the Edge (Blog) →

The hardware platform

A few words on the hardware platform. Camera sensors produce a lot more data than typical IoT sensors where you get a single numeric value each time. Therefore you may have to be a bit more careful when selecting the hardware platform for your edge processing, to make sure that it can cope with the data volumes. It also depends on the type of algorithm you want to use, machine learning models, especially convolutional neural networks (CNN), typically require significantly more processing power than basic vision algorithm. Still, even on a lightweight device such as the Raspberry Pi you can do basic vision processing.

Some scenarios might involve multiple cameras feeding into one compute platform where you can perform Edge Streaming Analytics on multiple camera feeds in one node. Make sure you dimension the hardware accordingly. For instance Intel has FPGA/GPU solutions for gateways that allows for more advanced video processing.

Also, some of the higher-end cameras have integrated processing capabilities and have the possibility to run 3rd party software, including Crosser Edge Node. For some use-cases, single video feeds for instance, that might be the better choice.

Automation and Integration based on Vision Inspection

The camera and algorithm will help you detect anomalies. But the full value is achieved when you can build automation and integration workflows based on the video streams.

The Crosser Platform is designed to be a streaming analytics platform but also a platform for automation and integration. Based on the result of your vision analysis the platform empowers you to:

  • Collect snapshots or recordings of the video stream and send to machine operators displays, on-premise systems or cloud providers
  • Send real-time triggers to machines to stop, start or initiate other actions
  • Send triggers into enterprise system to initiate work-orders or other business processes using API’s
  • Send notifications using SMS, email or other services like Slack etc

Pre-packaged Vision Inspection solutions vs DIY (do it yourself) using off-the shelf cameras and Edge Analytics Software

There are several pre-packaged Vision Inspection solutions in the market where one supplier is providing full solutions for specific use-cases. These are good when you need an end-to-end solution and want full end-to-end support of specialist vendors. The downside is high cost and a dependency on vendors for innovation.

DIY solutions requires you to do some more work yourself but this is becoming easier and easier with self-service platforms like Crosser in combination with open vision algorithms. If you chose this route you will have a horizontal solution that you can leverage for any use-case you can imagine. In addition, unbundling the tech-stack and using best-of-breed typically results in significant cost savings.

To know more about this solution and sign up for a demo or trial contact us here!

About the author

Goran Appelquist (Ph.D) | CTO

Göran has 20 years experience in leading technology teams. He’s the lead architect of our end-to-end solution and is extremely focused in securing the lowest possible Total Cost of Ownership for our customers.

"Hidden Lifecycle (employee) cost can account for 5-10 times the purchase price of software. Our goal is to offer a solution that automates and removes most of the tasks that is costly over the lifecycle.

My career started in the academic world where I got a PhD in physics by researching large scale data acquisition systems for physics experiments, such as the LHC at CERN. After leaving academia I have been working in several tech startups in different management positions over the last 20 years.

In most of these positions I have stood with one foot in the R&D team and another in the product/business teams. My passion is learning new technologies, use it to develop innovative products and explain the solutions to end users, technical or non-technical."

Close