Correlation filters tissue tracking with application to robotic minimally invasive surgery

Vision-based tissue tracking is a significant component for building efficient autonomous surgical robot system. While the methodology involves various challenges caused by occlusion, deformation and appearance changes.

We propose a novel correlation filter tissue tracking framework for minimally invasive surgery. Our model contains the innovative design of synthetic features, a bi-branch is exploited to enhance the response map. An incrementally learnt detector with the novel updating and trigger schemes is embedded to model the re-detection module for capturing the lost target.

Promising validation has been conducted on the publicly available tracking benchmark datasets, a surgical tissue tracking dataset based on publicly available Cholec80 dataset has also been developed to focus on the application in intra-operative scenes.

Our proposed framework meets the outstanding performance and surpasses the existing methods. The work demonstrates the feasibility to perform tissue tracking by taking advantage of the correlation filter.