Machine vision, also sometime referred to as computer vision, is an extremely important field of computer science that is likely to play a gigantic role in the direction of technology and society moving forward.
But while many people will focus on the software aspect of machine vision, it is all-too-easy to forget the equally important practical elements that will influence overall performance. One such example is the use of optical coatings and filtering: both of which will either extend or severely limit the possible applications for this highly exciting technology.
Computer Vision and Optics
Computer vision/machine vision is the ability of a computer to ‘see’. It does this by using a camera to create a digital image, and then analysing the data that is contained in that image. This is a technology that has been used for a long time, but in the last few years it has been rapidly increasing in importance. That’s because machine learning has enabled rapid improvements in this field, to the point that a computer can not only ‘see’ but also understand precisely what it is seeing: using this information to identify elements in a scene, or even to navigate in 3D space.
When you think of computer vision, you might think about robotics: specifically, robots that move through a room. This is one application to be sure, but others also include VR, facial recognition, digital assistants, data processing, social media, and much more.
VR for example uses computer vision in order to understand 3D space, thereby keeping the user safe while they enjoy their immersive experience, while also tracking their virtual movements to their real-world ones.
In order for all this to work, filtering is needed. These filters are created by doping glass materials with elements that can help to alter the absorption and transmission spectra.
The precise elements, or ‘dopants’, depend precisely on the wavelength that is desirable for the application.
The role of these filters is several fold. In some cases, filtering might be used in order to help protect the substrate underneath. For instance, a coating can prevent bright sun from damaging machinery and this could be important for a drone that is being flown in harsh weather conditions.
Likewise though, filtering can help to provide the first steps in the computational processes that allow machine learning to occur. When navigating through a virtual space for instance, a computer program will only need to look for contrast – which typically denotes an edge. The right filter can help to increase the contrast of the image, thereby making it easier to navigate the scene with less on-board processing necessary.
Another type of filter might be used in order to provide data not visible to the human eye. For example, an IR light can create a false color on a camera that can degrade the color reproduction and therefore many imaging cameras will use an IR-cut filter for the sensor.
Conversely, some technologies will use invisible light waves such as IR precisely because they can’t be seen by the human eye. An example is the ‘Leap Motion’ hand tracker.
Types of Coating
There are many types of coated filters used in this technology. Typically, coated filters are intended to offer sharper cut on and off transitions and higher transmissions. They are superior in these ways to other colored glass filters.
Every coated filter will go through a unique manufacturing process that ensures it meets performance targets. Wavelength-selective filters are manufactured using the deposition of dielectric layers added to the substrate. These have high and low refraction indices respectively and can combine to produce a range of desired results.
Surface quality and uniformity are extremely important factors when choosing the substrate as this can drastically impact on the performance and longevity of the coating.
There are a wide range of different types of filters, which include bandpass, longpass, shortpass, and notch filters. These each have specific blocking ranges. The explanation is in the name in each case, where the ‘long pass’ filter allows the longest wavelengths to pass through, blocking the shorter wavelengths. Short pass will block longer wavelengths and allow the shorter wavelengths to pass through. Bandpass filters block both longer and shorter wavelengths while only allowing a selected wavelength band in the middle to pass through. Notch filters will only allow wavelengths at either end of the spectrum to pass through while only blocking a selected wavelength band in the middle. Think of this as being like a cut-out or ‘notch’ in the middle of the signal.
Using these coated filters on cameras and other technologies is a relatively straightforward process, but the unique application of machine vision can introduce some unique difficulties. For example, these filters are designed for a specific Angle of Incidence (AOI) which is normally 0 degrees. This means that only light hitting the lens head-on will be blocked at the right wavelengths. This issue is particularly pronounced with the use of wider angle lenses – as is common in machine vision. Solutions can be applied however, such as moving the lens itself, or having multiple layers of glass. Alternatively, multiple lenses may be used rather than one larger one. Software correction can also help to reduce the noise.
Even with that limitation in mind however, coated filters will offer superior performance in most cases and is almost always preferred.
As with all things though, it is important to consider the precise application and other goals of the project. There are huge varieties of different types of machine vision, and using the right filter will depend on the environment, the goals, and the type of image analysis.
If you’re in need of any type of coating or filter contact Evaporated Coatings!