From Algorithm development to Industrial product

Challenge 

Light is an electromagnetic wave, with

• Amplitude perceived as brightness
• Wavelength perceived as color
• But the orientation of the
  wave oscillation is not visible

Properties of Polarization:

• All orientations equally present
  --> degree of polarization 0%
• Only one strong orientation
  --> degree of polarization 100%
• All orientations are present
  but at different strength
  --> partially polarized, e.g. 50%
• Dominant angle of polarization
  --> another degree of freedom

Where does it come from?

Most light sources (except lasers) are unpolarized

• Reflection gives polarized light, think of water surfaces, glass, beamsplitters surface inspection, ellipsometry
• Reflection at non-metallic surfaces like human skin, plants, plastics, object classification

• Nanostructures can create polarization in transmission and reflection, useful for polarization filters, carbon fiber inspection

Measuring Polarization

• Polarization has 3/4 degrees of freedom
  • Intensity of light
  • Degree of polarization
  • Angle of maximum polarization
  • (Circularity)

We can use ≥3 measurement with polarization filters for analysis
(and some math to obtain Stokes parameters)

But from there: 

• How to process the raw sensor data?
  • Different from RGB debayering
  • How do you calibrate measurements?
  • Achieving reliable data for measurements
• How to solve the applications
  • What is a good setup?
  • The user does not need polarization values
  • More processing is required to calssify
    industrial products for quality (good/bad)

Solution

Algorithm development with internal prototype camera

• Image processing for evaluating
  different modes/calculations
• Extending SciLab
  • Interfacing an industrial USB
    camera for live image capture
  • Interfacing a step motor controller
    for defined filter rotation

 Processing Stack for Prototype System

Results

Product Development: For integration into industrial inspection systems

• Use upcoming industrial cameras
  • 5 MPixel, single-shot imaging
  • High frame rate for inline inspection
• Provide the software as extension/plugin/.dll
  for machine vision software
• Calibrated measurements of retardation/ATN
• Measurement of magnitude and orientation of stress

From Scilab Code to a .dll

• Re-programm everything again
• Transfer into parallel C code
• Utilize parallel architectures
• Take into account:
  • Worst-case execution time
  • hard realtime applications

 Toolflow used

 Processing Stack for Commercial Solution