| This dissertation proposes an accurate, robust and unsupervised tracking system for tracking rolling leukocytes in vivo. First, two particle filter based tracking approaches, the Monte Carlo tracker and the affine transform invariant approach, are developed to track a single rolling leukocyte in vivo. In the Monte Carlo tracker, a specialized sampling scheme is proposed, which is able to generate efficient samples to overcome sample degeneracy without reducing sample diversity. To deal with the erratic motion of rolling leukocytes, an adaptive state transition model is applied. In addition, on the basis of the leukocyte image intensity properties, a weighting criterion is tailored, which mitigates the effect of the image clutter caused by the vessel wall. In the affine transform invariant approach, the affine transform invariant property is incorporated into the particle filter framework, which enables the developed method to accommodate linear affne transformations (translation, rotation, and/or scaling) of the target without the necessity of determining the exact transformations and the constituent parameters. Moreover, the affine transform invariant approach is more robust to image clutter and in tracking dark leukocytes. The incorporation of the flexible affne transform invariant property makes the affine transform invariant approach a very general tracking method that does not have the shape or size constraint on the target and can be applied widely, especially for tracking those targets that may rotate or may change their scales between frames. Compared with the snake tracker, which is well established for tracking a single rolling leukocyte in vivo, both the Monte Carlo tracker and the affine transform invariant approach exhibit superior performance in terms of the reduced localization error and the increased number of frames tracked.; To enhance tracking efficiency, a multi-leukocyte tracking system is developed, which has the ability to simultaneously track all the leukocytes in the field of view, to identify the new leukocytes in a frame, to automatically initialize the tracking process, to stop tracking leukocytes that exit the field of view, and to handle possible occlusion between leukocytes. (Abstract shortened by UMI.)... |
