Medical doctors can access and visualize the gastrointestinal tract through an endoscope.
However, existing endoscopes have several limitations such as inaccessibility to the entire
intestine and high discomfort. The wireless capsule endoscope, with external guidance for
interactive GI tract examination, tries to address some of these limitations. However, the
current techniques do not provide good localization of the endoscope capsule.
In this thesis, I propose a new wireless robotic capsule endoscope that uses a magnet
based localization and orientation system. Here, the capsule, which encloses a permanent
magnet, moves in a patient’s GI tract, and the magnet creates a magnetic field. The field
intensities at some spatial points are measured by magnetic sensors placed around the
patient’s body. We have derived the magnetic intensities as nonlinear functions of
magnet’s 3D location and 2D orientation parameters, which can be solved by minimizing
an objective error function. We have evaluated several optimization algorithms and found
that the Levenberg-Marquardt method provides superior performance in terms of
numerical accuracy and execution speed. Although a nonlinear method could be an
approach to the problem, it requires a good initial guess of the unknown parameters, and
its speed is low. Therefore, I propose a linear algorithm. With the data from five (or more)
3-axis magnetic sensors, this algorithm is realized through matrix computations. In most
cases, it results in correct solution if the matrix is not singular, and its execution time is
less than one-tenth that of the fastest nonlinear algorithm.
Finally, a real magnetic sensor array system has been built with sixteen Honeywell
3-axis AMR sensors. With a computer interface for 3D graphic and data display, real-time
tracking is realized, and the results are satisfactory with an average localization error of
about 3.3 mm and an orientation error of about 3.0º in case that the magnet moves within
the area of the sensor array. The magnet’s orientation misses the information of capsule’s
rotation along its own central axis, and hence a supplemental localization and orientation
method is proposed by an imaging technique. Applying an 8-point algorithm on the
images after distortion correction, we solve the 3D rotation and 2D transition parameters
of the camera and this rotation can be further decomposed into two sub-rotations, the
rotation of the camera’s central axis and the rotation around this axis.
Keywords: Localization and orientation, Robotic wireless capsule endoscope, Magnetic
tracking, Image processing and Computer vision.