Polarimetric Monocular Dense Mapping Using Relative Deep Depth Prior

This letter is concerned with polarimetric dense map reconstruction based on a polarization camera the help of relative depth information as prior. In general, imaging able to reveal about surface normal such azimuth and zenith angles, which can support development solutions problem reconstruction, especially in texture-poor regions. However, shape cues are ambiguous due two types polarized reflection (specular/diffuse). Although methods have been proposed address this issue, they either offline therefore not practical robotics applications, or use incomplete cues, leading sub-optimal performance. letter, we propose an online method that uses full available from camera. With our method, propagate sparse values both along perpendicular iso-depth contours. Through comprehensive experiments challenging image sequences, demonstrate significantly improve accuracy depthmap well increase its density, specially regions poor texture.