
There are only a limited number of physical quantities that can be measured directly with optical techniques; intensity, phase, and polarization, and a few more that can be measured directly from images i.e. deformations. All other quantities of interest are subject to estimation through models, preferably physically based models. Spatially, temporally and/or spectrally resolved information provide excellent information for fitting to models and thereby deduce quantities not directly measured, but often it is sufficient to collect information from discrete points in the space – time – frequency space. But this requires good reliable models. An example of the latter is automotive sensors where data economy and low cost is crucial for acceptance. The figure below shows the response of a forward looking two-wavelength sensor for road condition monitoring mounted on a truck when driving past different patches of dry, wet, icy and snowy asphalt.
Fig 1. Accuracy of an optical road surface sensor based classification method (below) compared to a friction coefficient measured by using a sliding weel (above). The marks show when the estimation is within (green circle) and outside (red cross) the lower and upper friction boundaries for the identified surface