A fast, low-cost, computer vision approach for tracking surgical tools

Abstract

Given the rise in surgeries performed with surgical robots and associated robotics research efforts, tool tracking methods have the potential to provide quantitative feedback concerning surgical performance and establish absolute tool tracking to help advance surgical robotics research. We have created a platform-agnostic method for low-cost tracking of surgical tool shafts in Cartesian space in near real time. We employ a joint Hough Transform - Geometric Constraint approach to locate the tool tips in the stereo camera channels independently. Cartesian coordinates are registered using a custom polynomial depth - disparity model. The algorithm was developed using a low-cost experimental webcam setup and evaluated using a da Vinci surgical endoscope. The algorithm was benchmarked for 3D tracking accuracy and computational speed. The system can locate the tool tip in 3D space with an average accuracy of 3.05 mm at 25.86 frames per second using the webcam setup. For the endoscope setup this algorithm has an average tracking accuracy of 8.68 mm in 3D and 1.88 mm in 2D with an average frame rate of 26.9 FPS. The algorithm also demonstrated successful tracking of tools using captured video from a real surgical procedure.