Real-time methods for long-term tissue feature tracking in endoscopic scenes

Abstract

Salient feature tracking for endoscopic images has been investigated in the past for 3D reconstruction of endoscopic scenes as well as tracking of tissue through a video sequence. Recent work in the field has shown success in acquiring dense salient feature profiling of the scene. However, there has been relatively little work in performing long-term feature tracking for capturing tissue deformation. In addition, real-time solutions for tracking tissue features result in sparse densities, rely on restrictive scene and camera assumptions, or are limited in feature distinctiveness. In this paper, we develop a novel framework to enable long-term tracking of image features. We implement two fast and robust feature algorithms, STAR and BRIEF, for application to endoscopic images. We show that we are able to acquire dense sets of salient features at real-time speeds, and are able to track their positions for long periods of time. © 2012 Springer-Verlag.