Irene Clemente, "Investigation and Implementation of a level-set based tracking system, incorporating contour prediction by means of different, collaborating prediction algorithms", 2008.

Object tracking is a challenging task in many computer vision applications such as automated surveillance systems, driver assistance and humanoid robots. A crucial step in object tracking is object segmentation, which is in charge of the detection of the object to track. Level set methods are a powerful approach for image segmentation. These methods are based on the iterative deformation of a surface, the level set function, which provides the contour of the object. Nevertheless, the main drawback of these methods, especially in real-time systems, is their computational cost when a large number of iterations is required. Different prediction tracking algorithms, which detect the object movements along the image sequences, are investigated and implemented in this thesis. They reduce the computational cost of the process by means of a predicted level-set function for the next frame of the sequence. From the evaluation of the prediction algorithms, it is concluded that each algorithm depending on the field of application leads to different levels of reliability and confidence in the prediction. A collaboration between the different algorithms provides a more robust prediction, driving to a much faster convergence of the level set. The segmentation task presented in this work is integrated into an object tracking framework. This detection module segments the target object in each frame of a video sequence, serving as a template of the object for the following processing tracking steps. Experimental results demonstrate and confirm the good performance, robustness, and efficiency of the collaborating prediction algorithms in a level set-based tracking system. Several different sequences in this thesis are tested, in which the computational cost of the level set segmentation is efficiently reduced and the drift produced by the template adaptation mode of other tracking system is overcome.