Tracking and shape estimation of flexible robots that
navigate through the human anatomy are prerequisites to safe
intracorporeal control. Despite extensive research in kinematic
and dynamic modelling, inaccuracies and shape deformation of
the robot due to unknown loads and collisions with the anatomy
make shape sensing important for intra-operative navigation. To
address this issue, vision-based solutions have been explored. The
task of
2
D tracking and
3
D shape reconstruction of flexible robots
as they reach deep-seated anatomical locations is challenging,
since the image acquisition techniques usually suffer from low
signal-to-noise ratio (SNR) or slow temporal responses. Moreover,
tracking and shape estimation are thus far treated independently
despite their coupled relationship. This paper aims to address
tracking and shape estimation in a unified framework based
on Markov Random Fields (MRF). By using concentric tube
robots as an example, the proposed algorithm fuses information
extracted from standard monoplane X-ray fluoroscopy with the
kinematics model to achieve joint
2
D tracking and
3
D shape
estimation in realistic clinical scenarios. Detailed performance
analyses of the results demonstrate the accuracy of the method
for both tracking and shape reconstruction.