Towards live monocular 3D laparoscopy using shading and specularity information

Abstract

We present steps toward the first real-time system for computing and visualising 3D surfaces viewed in live monocular laparoscopy video. Our method is based on estimating 3D shape using shading and specularity information, and seeks to push current Shape from Shading (SfS) boundaries towards practical, reliable reconstruction. We present an accurate method to model any laparoscope's light source, and a highly-parallelised SfS algorithm that outperforms the fastest current method. We give details of its GPU implementation that facilitates realtime performance of an average frame-rate of 23fps. Our system also incorporates live 3D visualisation with virtual stereoscopic synthesis. We have evaluated using real laparoscopic data with ground-truth, and we present the successful in-vivo reconstruction of the human uterus. We however draw the conclusion that the shading cue alone is insufficient to reliably handle arbitrary laparoscopic images. © 2012 Springer-Verlag.