Robust feature matching for VSLAM in non-rigid scenes

Minimally invasive surgery (MIS) has a wide range of applications in the medical field. Its emergence has brought many benefits to patients but has also brought great challenges to surgeons, requiring doctors with extensive surgical experience. The visual simultaneous localization and mapping (VSLAM) method can track the position of the endoscope during the operation and complete the 3D reconstruction of the scene, which can provide significant help to the surgeon through surgical navigation. Feature matching is an essential link in VSLAM, and the correct correspondence of feature points between two images determines the overall accuracy of VSLAM surgical navigation tasks. Aiming at the problems of nonrigid transformation and noise in medical image feature matching jobs, we propose a robust feature matching algorithm that can handle the effects of soft tissue deformation. This paper is validated using publicly available medical endoscopy datasets and compared with other state-of-the-art methods. The experimental results show that our method outperforms different approaches and has better performance in surgical scenarios.