Filtering for Closed Curves

Shape analysis;Tracking;Particle filtering

Rathi, Yogesh ; Electrical and Computer Engineering

University:Georgia Institute of Technology

Department:Electrical and Computer Engineering

关键词: Shape analysis;    Tracking;    Particle filtering;   

Others  :  https://smartech.gatech.edu/bitstream/1853/13996/1/Rathi_Yogesh_N_200612_phd.pdf

美国|英语

来源: SMARTech Repository

【 摘 要 】

This thesis deals with the problem of tracking highly deformableobjects in the presence of noise, clutter and occlusions. Thecontributions of this thesis are threefold:A novel technique is proposed to perform filtering onan infinite dimensional space of curves for the purpose of trackingdeforming objects. The algorithm combines the advantages of particlefilter and geometric active contours to track deformable objects inthe presence of noise and clutter.Shape information is quite useful in tracking deformableobjects, especially if the objects under consideration get partiallyoccluded. A nonlinear technique to perform shape analysis, calledkernelized locally linear embedding, is proposed. Furthermore, a newalgebraic method is proposed to compute the pre-image of theprojection in the context of kernel PCA. This is further utilized ina parametric method to perform segmentation of medical images in thekernel PCA basis.The above mentioned shape learning methods are then incorporated into ageneralized tracking algorithm to provide dynamic shape prior fortracking highly deformable objects. The tracker can also model imageinformation like intensity moments or the output of a featuredetector and can handle vector-valued (color) images.