Deconvolution-Based Structured Light System with Geometrically Plausible Regularization

This paper presents a new deconvolution-based energy formulation for segmenting the image of stripe-based patterns projected by structured light systems. Our framework features an explicit modeling of the blurring introduced by the lens of a structured light system. This allows a significant improvement when working out of focus, a situation which occurs when performing depth measurement. The proposed iterative algorithm includes two steps: a deconvolution and a segmentation. For both steps, a geometrically plausible regularization term is used. It considers the projected displacement induced by the camera, projector and scene configuration. We validate our method using real imagery acquired using off-the-shelf equipment.