Tracking of the tongue in three dimensions using a visual recording system

Hageman, T.A.G. (2013) Tracking of the tongue in three dimensions using a visual recording system.

Abstract:Oral cancer is a disease which can signi�cantly a�ect one's oral abilities, including speech, food transport, chewing and swallowing. There are several treatment methods, but the choice of treatment is determined by subjective means. The Dynamic Virtual Surgery project aims to develop a system which allows the study of post-operative function loss by pre-operative simulations. For tongue cancers, this means that a good patientspeci �c tongue model must be constructed. The training process of such a model involves tracking the tongue shape in three dimensions during an operation, while simultaneously acquiring EMG data of the tongue. This thesis involved the development of a system able to track the shape of the tongue in three dimensions. The chosen approach uses three cameras placed in front of the patient, converting the information of multiple cameras to a 3D-representation. Di�culties in this process include the smooth, moist and occlusion-prone environment of the tongue. Markers are stuck to speci�c locations on the tongue in order to de�ne high-contrast landmarks, a process involving the use of non-toxic materials. Reproducibility of marker locations is guaranteed to a certain extend by preparing the layout beforehand on a exible bandage which is stuck to the tongue. The tracking algorithm works o�ine and involves several important processes. Template matching is used to �nd the marker coordinates in the recorded frames. An outlier correction algorithm is then run to correct for measurement errors. Finally, a Kalman �lter is used in order to track the state of the tongue, directly transforming the 2D-measurements to a 3D-representation. To deal with outliers, a method is proposed involving the use of a principal component (PCA) model. This model, based on the 3D marker locations, allows reduction of dimensionality and does not allow physically impossible tongue states. A method similar to RANSAC, involving hypothesis generation and -testing on multiple subsets of measured marker coordinates in the recorded frames, determines which PCA components are most likely for the current situation, which can then be used to correct the outliers. The method proves to be working and provides results with a 3D-accuracy down to sub-millimeter level. In the case of occluded markers, accuracy drops but still remains below two millimeters.
Item Type:Essay (Master)
Faculty:EEMCS: Electrical Engineering, Mathematics and Computer Science
Subject:50 technical science in general
Programme:Electrical Engineering MSc (60353)
Link to this item:
Export this item as:BibTeX
HTML Citation
Reference Manager


Repository Staff Only: item control page