Key points are not available for this paper at this time.
Conventional binocular stereo vision is limited by the photometric consistency assumption, which often fails in regions with strong specular reflections or sparse texture. Although polarization information can provide additional geometric constraints, in high dynamic range scenes, the degree of linear polarization may be abnormally reduced due to sensor clipping, rather than true depolarization. In this work, this issue is analyzed, and a depth estimation method is proposed based on a division-of-focal-plane polarization camera that simultaneously considers polarization reliability and intensity validity. By modeling polarization reliability as a continuous function, smooth fusion with intensity-based depth is achieved. Experimental results show that the proposed method improves depth completeness and consistency in complex scenes.
Zhu et al. (Fri,) studied this question.