ECCV 2016 Daily - Thursday

Silvia Tozza 22 ECCV Daily : Thursday Presentations the first monocular, passive shape-from-x technique that enables depth estimation with only a single, uncalibrated illumination condition, which is a significant novelty. Their goal is to estimate the surface depth of an object using the polarisation properties of the reflected light. These properties are contained in a 3-channel polarisation image which can, in principle, be captured in one shot. Their approach needs only a single polarisation image. They use polarisation constraints and shading constraints, expressing them directly in terms of the surface depth as linear equations. In this way, they simply solve a large, sparse system of linear equations to compute a depth map. Silvia Tozza is a postdoc at the Department of Mathematics of the Sapienza - Università di Roma in Italy. She accepted to talk to us about the model she is presenting at ECCV2016 . What she told us is that, at the moment, polarisation is a very underexploited cue for the 3D reconstruction of objects . This is surprising, because the shape- from-polarisation cue has some nice properties. For example, it does not depend on the position and/or the number of lights but only on the viewer position. We think that our work could be a useful tool since solving for depth is very efficient and it works well on glossy objects, including in uncontrolled, outdoor illumination. To her knowledge, their method is Linear depth estimation from an uncalibrated, monocular polarisation image