Dense Surface Reconstruction With Shadows in MIS

Three-dimensional reconstruction of internal organ surfaces provides useful information for better control and guidance of the operations of surgical tools for minimally invasive surgery (MIS). The current reconstruction techniques using stereo cameras are still challenging due to the difficulties in correspondence matching in MIS, since there is very limited texture but significant specular reflection on organ surfaces. This paper proposes a new approach to overcome the problem by introducing weakly structured light actively casting surgical tool shadows on organ surfaces. The contribution of this paper is twofold: first, we propose a robust approach to extract shadow edges from a sequence of shadowed images; second, we develop a novel field surface interpolation (FSI) approach to obtain an accurate and dense disparity map. Our approach does not rely on texture information and is able to reconstruct accurate 3-D information by exploiting shadows from surgical tools. One advantage is that the point correspondences are directly calculated and no explicit stereo matching is required, which ensures the efficiency of the method. Another advantage is the minimum hardware requirement because only stereo cameras and a separated single-point light source are required. We evaluated the proposed approach using both phantom models and ex vivo images. Based on the experimental results, we achieved the precision of the recovered 3-D surfaces within 0.7 mm for phantom models and 1.2 mm for ex vivo images. The comparison of disparity maps indicates that with the addition of shadows, the proposed method significantly outperforms the state-of-the-art stereo algorithms for MIS.