Multimodal Registration Using Stereo Imaging and Contact Sensing

Abstract

Registration plays an important role in robot assisted minimally invasive surgeries, by localizing the tool-tip onto the preoperative model of the anatomy. Contact-based point measurements have been typically used to perform registration. However, a lot of time is spent in moving the robot tool to obtain these measurements. On the other hand, stereo imaging quickly provides thousands of point measurements, but these measurements are affected by noise in the measurements as well as artifacts such as occlusions and specular reflections. In this work, we use both stereo and contact measurements in a complementary manner to obtain fast and accurate registration estimates. We show that stereo measurements provide fast and approximate registration estimates, which can then be refined with as few as 10 contact measurements. We validate our approach with experiments performed using a daVinci surgical robot on femur bone and silicone phantom tissue. In comparison with the popular registration method iterative closest point, our approach is faster by an order of magnitude.