10.5. Limitations

The implementation of the raytracer is performed based on object-oriented programming (OOP) models. The inheritance and polymorphism principles of OOP allow for efficiently replacing, adapting or enhancing the scattering models for new materials or surfaces. Currently, apart from a few auxiliary generic classes, only one scattering class (TLSISroughPol, see section 7.9) is used to model materials. While the integrated features allow modelling a wide range of opaque or transparent, coated or uncoated and rough or flat surfaces, it will still be necessary to add new models for modelling practically relevant surfaces realistically.
The statistical approaches used to model the surface’s roughness as well as the implemented diffuse subsurface reflection model are based on the assumption of randomly distributed surface orientations or scattering particles. Consequently, the implemented models are unable to correctly describe scattering processes on surfaces that exhibit significant regular structures on the microscopic or nanoscopic scale.
The methods to derive new scattering classes can essentially be divided into three types, which can be referred to as white-, grey- and black-box approaches. The white box approach can be applied to model micro-structured surfaces, like coarse textiles or perforated metals. In this approach, light scattering is explicitly modelled based on a microscopic three-dimensional model of the surface. Apart from the three-dimensional model, these models generally require limited empirical data to calibrate their scattering properties to real-world materials. However, the validity of the approach is limited to geometric structures of larger scales, as diffraction becomes more significant when the relevant apertures are small.
To model materials dominated by smaller-scale structures, appropriate statistical models can be applied to model the distribution and orientation of scattering events. Again, measured data can be used to calibrate the model parameters. Finally, in the black-box approach, scattering classes can be derived based on measured data only. The information from angular measurement samples (BSDF) can be used to model the related scattering events directly. However, polarisation information is generally disregarded in this approach, as it is usually not contained in BSDF measurements. Beyond the need for additional scattering classes, the demand for suitable material data can pose an issue. The topic is discussed in detail above (section 10.3).
Another limitation is not related to the optical modelling but to the SIOP operator, which is used to pass the raytracing information to other applications. According to its definition, the output of the operator is limited to one dimension. Therefore, the concept cannot readily be used to replace the currently applied Klems matrices. The directionally resolved output information is a fundamental requirement for most lighting applications. It is, however, not relevant for the energetic simulations this work focuses on. However, since the RadiCal method intrinsically covers the visible spectrum of light, it may be well worthwhile to consider its application in this field. The RadiCal raytracer can be used to generate Klems matrices, allowing its immediate use for lighting applications. While this approach has the advantage of being compatible with the existing methods, it would be desirable to exploit the advantages of the SIOP definition in the lighting field as well. One potential approach involves generating a series of SIOPs covering the hemisphere for the outgoing light with the desired resolution. However, this would still correspond to a discrete distribution on the output side. A more complex solution would involve the development of a more complex, multidimensional representation still based on spherical harmonics. E.g., the resulting function values of the current approach could provide coefficients for another spherical harmonics expansion describing the directional distribution of the outgoing light.