Machine vision is a subfield of artificial intelligence and computer science that helps machines understand and interpret the world visually with sensors or cameras. This involves using various algorithms and techniques to analyze digital images or videos and extract useful information from them.
Machine vision systems are composed of different components apart from cameras and sensors, such as image processing hardware and software, and a way to display results or take action based on what the machine has “seen”. The camera captures images or videos, and then specialized hardware and software analyze them to perform various tasks.
Machine vision, therefore, teaches machines to see and recognize objects the same way humans do (but much faster and more efficiently). Consequently, they use different techniques like feature detection and matching, template matching, and neural networks based on deep learning to classify objects in images or videos.
On the other hand, embedded vision systems refer to using computer vision algorithms and techniques in embedded systems.
Embedded vision systems function as part of a bigger device and usually have a specific task within a more comprehensive system. Some systems can easily implement multiple embedded vision systems to perform the tasks necessary for the application.
Because embedded vision systems usually have a specific task within a more extensive system, they combine computer vision techniques with small, resource-constrained computing devices. Embedded vision systems aim to allow these devices to see and interpret the world around them, allowing them to perform tasks that previously required human vision and understanding, much like machine vision systems. Embedded vision systems are, therefore, usually physically smaller and simpler to integrate than machine vision systems.
For all intents and purposes, when it comes to optical filters, though, the term embedded vision is more or less interchangeable with machine vision. This is also the main reason we have categorized vision systems under one category (machine vision) since embedded vision systems are also machines (it’s in the name) with a similar purpose as other machine vision systems.
As we’ve demonstrated above, machine vision and embedded vision are more or less interchangeable technologies.
Whereas machine vision and embedded vision are systems engineering disciplines, computer vision is a field of study based on artificial intelligence and computer science. Therefore, computer vision is actually a broader definition that encompasses many different applications. In contrast, machine vision and embedded vision are more specific applications of computer vision in industrial automation and embedded systems.
By drawing this line in the sand, we can conclude that computer vision is not the technically correct terminology to describe the applications we focus on here at PSC. Therefore, we’ll only focus on machine vision and embedded vision applications in the following examples.
While both machine vision and embedded vision involve computer vision, there can be slight differences in application and deployment. Machine vision systems are typically used in industrial and manufacturing settings, where accuracy and precision are critical. These systems often require high-performance computing hardware and are deployed in fixed locations.
On the other hand, embedded vision systems are designed to be deployed on small, mobile, or low-cost devices. They are often used in consumer electronics, healthcare, automotive, and other industries. These systems must be optimized for power consumption, cost, and performance and often require specialized hardware platforms to achieve this.
There are many similarities when choosing optical filters for these types of applications.
As we have illustrated, the line between when something is technically a machine vision or embedded vision application is often blurred. The requirements for the optical filter for an embedded vision and machine vision application could even be the same in some scenarios.
Therefore, the choice of optical filter relies more on what is required of the vision system in question and what environment it will be placed in – just like with any other kind of optical filter. We still consider the five areas to consider when choosing an optical filter solution, whether it’s called a machine vision application or an embedded vision application.
For this reason, it’s hard to conclusively say that optical filters for machine vision applications need one set of features and performance, and embedded vision systems need another set of specifications.
The same questions when choosing an optical filter for any other kind of application are still crucial to consider, such as requirements for image quality, contrast enhancement, durability, design, etc.
In summary, machine vision and embedded vision are two highly similar fields with slightly different focuses and applications – and computer vision is a more general term used to describe a field of study.
The similarities are so many, especially in regard to choosing optical filters for their devices, that we categorize all these kinds of applications as machine vision.
The main difference when you read about these technologies elsewhere is that while machine vision usually focuses on industrial automation and quality control, embedded vision implements computer vision into small, resource-constrained devices. These vision systems are used in a large variety of applications – maybe even in combination.