As machine vision systems mature, they’ll integrate with other emerging technologies, or technologies that have been in the landscape for a while but are now coming into their own. One such technology is AI.
The use of AI in machine vision systems is a positive development. AI can enhance machine vision systems’ effectiveness, which widens the range of machine vision systems’ applications. Learn what AI is suitable for, when it makes sense to implement it, and how to get started.
The Best Applications for AI in Machine Vision Systems
AI is a great fit for situations in which you need complex image analysis. You can train it to recognize particular objects, so it will only focus on those items when it’s working. AI can enhance the effectiveness of traditional machine vision systems. There are two other reasons to use AI in machine vision: precision and efficiency. AI can learn what to look for thanks to rigorous training, and this technology can pinpoint what it’s looking for faster and more precisely than humans.
The case studies below illustrate how companies are using AI in their machine vision systems.
Case Studies
Teledyne and Intelligent Traffic Systems
Teledyne developed an intelligent traffic system (ITS) that would identify vehicles as they drove past a camera with the goal of catching drivers who didn’t pay tolls for Canadian provinces. The engineers on the project had a machine vision system in place that relied upon optical character recognition (OCR), although the system’s accuracy was only 75%.
The engineers decided to integrate AI into their project. They trained AI to identify vehicles passing the camera at high speeds. The addition of AI was especially helpful because the engineers taught it to identify vehicles in various weather conditions, such as sun, snow, and rain. Moreover, the AI model was able to identify a variety of license plates from various provinces. Some of those license plates were too faded for a human eye to read at high speeds. Thanks to AI, the ITS was able to reach 95% accuracy.
Tesla and Self-Driving Cars
While self-driving cars aren’t widespread yet, Tesla has taken one step forward towards increasing their safety.
The electric carmaker has been working on an improved vision-based machine learning model which can identify objects. Engineers are also teaching the system to understand concepts such as position, velocity, and how those objects accelerate. The system works by tracking objects over time and uses “thresholding” to ensure those objects exist.
“Thresholding” means the system compares objects it’s been tracking over time to those appearing before its sensors. If those objects appear continuously, they meet the threshold for detection. Establishing a threshold is helpful, as it helps the system differentiate between real objects impermanent visual phenomenon such as smoke or fog.
Pleora Technologies Inc. and Quality Control
To err is human, which becomes a problem in the field of quality control. Many quality control activities such as inspection are still carried out by humans.
Pleora Technologies Inc., a Canadian company specializing in real-time connectivity software and hardware solutions for machine vision device and system manufacturers, aims to change that. Its Vaira platform integrates machine vision cameras, lighting and processing software with AI-based inspection and traceability apps.
The Vaira platform assists humans in quality control processes. Humans can become tired or distracted, which leads to mistakes. AI-enabled systems, on the other hand, don’t face those problems. With the platform, users can create customized apps which leverage machine vision and AI to enhance the visual inspection process and ensure it’s consistent, reliable and trackable.
When Does It Make Sense to Implement AI in Machine Vision Systems?
The temptation to integrate AI into machine vision systems might be strong. After all, no one wants to be left behind the competition.
However, there are three compelling reasons not to integrate AI into your machine vision system:
- The tasks the system performs involve clearly defined parameters.
- The parts involved are nearly completely uniform.
- The machine vision system reads basic bar codes.
A traditional machine vision system can handle those conditions quite well on its own. The conditions outlined above are simple and standard.
However, conditions aren’t always simple or standard. AI adds value when there are high variability and complex pattern recognition involved. The Teledyne ITS case study referenced earlier is a good example; AI was able to identify a variety of license plates from various Canadian provinces. Those license plates weren’t all of uniform quality; some of them were faded and difficult to read.
How to Get Started with AI in Machine Vision Systems
If you’re integrating AI into your machine vision systems, there are some components you’ll need to put into place:
- Computers with high processing power. AI requires extensive compute resources to work properly. Moreover, you don’t want to think about your current needs; you have to consider future needs, too. Choosing more advanced processors can position you for success in the years ahead.
- Comprehensive support for a variety of inputs and outputs. Your systems will be ingesting data from several systems and transmitting data, so make sure the hardware and software can handle those volumes of information.
- Secure high-throughput networking. With the proliferation of sensors, your machine vision system will need secure networking capabilities to ingest all that data.
- Safe power supply and intelligent power management. Devices connected to the machine vision and AI systems will require electricity as well as the ability to manage that power consumption.
- Compliance with industry standards. The hardware and software you choose must comply with industry standards for optimal performance and functionality.
- Durability. For implementations in industrial or manufacturing environments, the components must withstand dirt, dust, debris, vibrations, or other harsh conditions.
- The ability to scale. The choices you make now will affect you in the future. By choosing hardware and software that will grow with you, you’ll be able to grow going forward.
The Future of AI in Machine Vision Systems
Integrating AI into machine vision systems won’t make sense for everyone at present. There are still many use cases for which traditional machine vision systems make perfect sense. That being said, there’s a growing body of applications for AI in machine vision. It can increase precision and accuracy in cases of varying uniformity and highly complex images. Even if you don’t need it now, keep a lookout for future developments in AI; there might be a use case waiting for you in the future.
