Polarization Cameras in Machine Vision
Standard image sensors measure two things: how much light is hitting the pixel (intensity) and, if using a Bayer filter, what color it is (wavelength). However, standard silicon is completely blind to a third fundamental property of light: its polarization state. Polarization cameras and imaging systems equipped with polarization filters capture this missing data, measuring the exact angle at which light waves are vibrating. This allows vision systems to suppress specular reflections that interfere with inspection, increase the contrast of certain surface features, and improve detection of subtle surface defects and stress patterns in transparent plastics.
How it works: The on-sensor micro-polarizer array
Historically, machine vision systems captured this data using more complex methods, such as rotating a polarizing filter in front of the lens to capture multiple sequential images, or splitting light across multiple sensors using beam-splitting optics. The development of image sensors with integrated micro-polarizer arrays, such as Sony's Polarsens™ sensors, made modern polarization cameras practical for industrial use.
To capture polarization data in a single frame, sensor manufacturers bond a wire-grid polarizer directly to the silicon over the pixel array. This architecture is very similar to a color camera's Color Filter Array (CFA). The grid is divided into repeating 2x2 blocks. Each of the four pixels in a block is covered by a filter oriented at a different angle: 0°, 45°, 90°, and 135°.
When light hits this 2x2 block, the four pixels measure the light's intensity at those four specific angles simultaneously. The machine vision software then uses mathematical algorithms to calculate Stokes parameters (four values that together describe the full polarization state of a light wave) to determine the exact polarization of the light bouncing off the target.
What can vision software do with polarization data?
Because the camera captures the full polarization state in a single frame, the software can extract multiple different types of images from the same raw data:
-
Degree of Linear Polarization (DoLP): This calculates what percentage of the incoming light is polarized versus unpolarized. Highly reflective surfaces (like metal or glass) strongly polarize light, while matte surfaces diffuse it. DoLP images are exceptional for segmenting different materials that look identical in a standard grayscale image.
-
Angle of Linear Polarization (AoLP): This calculates the specific angle of the polarized light and assigns a false color (pseudocolor) to it. The colors represent physical angles rather than wavelengths. It is primarily used to detect 3D surface orientations and microscopic scratches on flat, uniform materials.
-
Virtual Glare Reduction (Reflection Removal): The software uses the polarization data to mathematically identify and suppress the strongly polarized component of the light causing the glare. By retaining the unpolarized component, it yields a perfectly clean image of the actual surface below.
Decision matrix for polarization applications
System integrators specify polarization cameras when standard optical tricks-like adjusting the angle of an LED light-fail to solve extreme contrast or reflection issues.
|
Application |
Engineering Challenge |
How Polarization Solves It |
|
Reading through plastic wrap |
Shrink wrap creates severe, unpredictable white glare that blinds standard cameras to the barcode beneath it. |
The software mathematically subtracts the glare (the polarized reflection), revealing the unpolarized light bouncing off the barcode. |
|
Glass and plastic stress testing |
Transparent materials undergo internal stress during manufacturing (birefringence), which weakens the part but is invisible to the naked eye. |
When placed in front of a polarized backlight, the internal stress alters the angle of the light, appearing as bright, colorful strain patterns in the AoLP image. |
|
Carbon fiber inspection |
Carbon fiber weaves are highly reflective, black, and geometrically complex, making standard edge detection nearly impossible. |
Polarization ignores the absolute brightness and isolates the physical direction of the carbon fibers, making the weave pattern clearly visible. |
The physical tradeoffs of on-sensor polarization
Just like the Bayer filter on a color camera, placing a physical grid over the silicon introduces engineering penalties that must be accounted for in your system design.
1. A significant reduction in light sensitivity
A polarizing filter, by definition, blocks light waves that are not perfectly aligned with its grid. Even before accounting for the physical wire grid blocking some photons, a polarizer inherently cuts the transmission of unpolarized ambient light by roughly 50%. This must be compensated for by increasing illumination intensity or using longer exposure times.
2. Reduced spatial resolution
Because the sensor relies on a 2x2 pixel grid to calculate a single polarization data point, effective spatial resolution is reduced, similar to the interpolation tradeoffs seen in Bayer color sensors.
Frequently asked questions
Yes, and that is a common technique, but it has severe limitations. A standard lens filter only blocks one specific angle of light globally. If your part has complex, curved geometry, the glare will reflect at multiple angles simultaneously; a lens filter might fix the glare on the top of the part but leave the glare on the sides. A polarization camera captures all angles at once, allowing the software to dynamically remove glare from complex 3D surfaces.
It depends on the application. For reflection removal and glare reduction (like reading through shrink wrap), standard unpolarized white light works perfectly, because the shiny surface creates the polarization you are trying to remove. However, for detecting internal material stress (birefringence), you must illuminate the transparent part from behind using a specialized polarized backlight.
Generally, no. The micro-polarizer array takes the physical place where a Bayer color filter would normally sit. Most industrial polarization cameras are built on monochrome sensors.