Abstract:

Polarization image simulation technology is a key means to break through the bottleneck of polarization data acquisition and promote the development of polarization vision. This study systematically reviewed three evolutionary paradigms of this technology: Physical mechanism simulation, based on the polarization bidirectional reflectance distribution function and polarization ray tracing, strictly solves polarization light transmission, which has high interpretability and credibility, but it is computationally complex and lacks visual realism. Data-driven simulation, using models like neural radiance fields to learn polarization appearance from data, has high generation efficiency and visual fidelity but weaker physical consistency and interpretability. Physics-data fusion simulation embeds physical constraints into data-driven models through differentiable rendering, balancing physical correctness and visual realism. This study summarized representative works of the three simulation paradigms, compared and analyzed the simulation performance and applicable scenarios of different paradigms, reviewed related datasets and simulation evaluation metrics, and discussed the current challenges and future development directions of the three paradigms.

Key words: polarization imaging, physical modeling, data drive, physics-data fusion simulation