LARK: A Low-Cost, Accurate, Occlusion-Resilient, Kalman Filter-Assisted Tracking System for Image-Guided Surgery

Abstract Image-guided surgery (IGS) depends on accurate tracking of surgical instruments to provide real-time navigation relative to critical anatomical structures. Most contemporary IGS systems utilize stereo infrared cameras to track fiducial markers attached to surgical instruments. However, such commercial solutions are prone to occlusion and remain cost-prohibitive for many healthcare settings. This work demonstrates that a multi-camera tracking system using commodity hardware can achieve sub-millimeter accuracy while addressing fundamental limitations of existing commercial solutions. Our system employs a ring-shaped fixture equipped with multiple RGB cameras mounted on a surgical light. We evaluate two tracking methods with adaptive Kalman filtering: multi-view monocular pose fusion and multi-view triangulation. Both methods are assessed under varying occlusion levels and compared against a gold-standard stereo infrared camera system. Results show our system achieves median target registration errors of 0.64 mm for static point tracking and 0.73 mm for dynamic trajectory tracking, with accuracy degrading gracefully under occlusion. Compared to existing solutions, our open source system offers a cost-effective, robust alternative for resource-constrained environments. All designs, code, and data are available at https://nist.mni.mcgill.ca/. Disclaimer: all links will be live after acceptance.