Simultaneous estimation of feature correspondence and stereo object pose with application to ultrasound augmented robotic laparoscopy

Abstract

In-situ visualization of ultrasound in robot-assisted surgery requires robust, real-time computation of the pose of the intra-corporeal ultrasound (US) probe with respect to the stereo-laparoscopic camera. Image based, intrinsic methods of computing this relative pose need to overcome challenges due to irregular illumination, partial feature occlusion and clutter that are unavoidable in practical robotic-laparoscopy. In this paper, we extend a state-of-the-art simultaneous monocular pose and correspondence estimation framework to a stereo imaging model. The method is robust to partial feature occlusion and clutter, and does not require explicit feature matching. Through exhaustive experiments, we demonstrate that in terms of accuracy, the proposed method outperforms the conventional stereo pose estimation approach and the state-of-theart monocular camera-based method. Both quantitative and qualitative results are presented.