Use this URL to cite or link to this record in EThOS: https://ethos.bl.uk/OrderDetails.do?uin=uk.bl.ethos.747606
Title: Visual tracking in robotic minimally invasive surgery
Author: Du, X.
ISNI:       0000 0004 7231 8122
Awarding Body: UCL (University College London)
Current Institution: University College London (University of London)
Date of Award: 2018
Availability of Full Text:
Access from EThOS:
Full text unavailable from EThOS. Please try the link below.
Access from Institution:
Abstract:
Intra-operative imaging and robotics are some of the technologies driving forward better and more effective minimally invasive surgical procedures. To advance surgical practice and capabilities further, one of the key requirements for computationally enhanced interventions is to know how instruments and tissues move during the operation. While endoscopic video captures motion, the complex appearance dynamic effects of surgical scenes are challenging for computer vision algorithms to handle with robustness. Tackling both tissue and instrument motion estimation, this thesis proposes a combined non-rigid surface deformation estimation method to track tissue surfaces robustly and in conditions with poor illumination. For instrument tracking, a keypoint based 2D tracker that relies on the Generalized Hough Transform is developed to initialize a 3D tracker in order to robustly track surgical instruments through long sequences that contain complex motions. To handle appearance changes and occlusion a patch-based adaptive weighting with segmentation and scale tracking framework is developed. It takes a tracking-by-detection approach and a segmentation model is used to assigns weights to template patches in order to suppress back- ground information. The performance of the method is thoroughly evaluated showing that without any offline-training, the tracker works well even in complex environments. Finally, the thesis proposes a novel 2D articulated instrument pose estimation framework, which includes detection-regression fully convolutional network and a multiple instrument parsing component. The framework achieves compelling performance and illustrates interesting properties includ- ing transfer between different instrument types and between ex vivo and in vivo data. In summary, the thesis advances the state-of-the art in visual tracking for surgical applications for both tissue and instrument motion estimation. It contributes to developing the technological capability of full surgical scene understanding from endoscopic video.
Supervisor: Stoyanov, D. Sponsor: Not available
Qualification Name: Thesis (Ph.D.) Qualification Level: Doctoral
EThOS ID: uk.bl.ethos.747606  DOI: Not available
Share: