<?xml version="1.0" encoding="ISO-8859-1"?>
	<metadata ReferenceType="Conference Proceedings">
		<site>sibgrapi.sid.inpe.br 802</site>
		<holdercode>{ibi 8JMKD3MGPEW34M/46T9EHH}</holdercode>
		<lastupdate>2008: sid.inpe.br/banon/2001/ administrator</lastupdate>
		<metadatalastupdate>2022: sid.inpe.br/banon/2001/ administrator {D 2008}</metadatalastupdate>
		<title>A Backmapping Approach for Graph-based Object Tracking</title>
		<format>Printed, On-line.</format>
		<size>9750 KiB</size>
		<author>Paixão, Thiago Meireles,</author>
		<author>Graciano, Ana Beatriz V.,</author>
		<author>Jr, Roberto M. Cesar,</author>
		<author>Jr, Roberto Hirata,</author>
		<affiliation>Instituto de Matemática e Estatística - USP</affiliation>
		<affiliation>Instituto de Matemática e Estatística - USP</affiliation>
		<affiliation>Instituto de Matemática e Estatística - USP</affiliation>
		<affiliation>Instituto de Matemática e Estatística - USP</affiliation>
		<editor>Jung, Cláudio Rosito,</editor>
		<editor>Walter, Marcelo,</editor>
		<conferencename>Brazilian Symposium on Computer Graphics and Image Processing, 21 (SIBGRAPI)</conferencename>
		<conferencelocation>Campo Grande, MS, Brazil</conferencelocation>
		<date>12-15 Oct. 2008</date>
		<publisher>IEEE Computer Society</publisher>
		<publisheraddress>Los Alamitos</publisheraddress>
		<tertiarytype>Full Paper</tertiarytype>
		<keywords>Digital Image Processing, Graph-based Segmentation, Object Tracking.</keywords>
		<abstract>Model-based methods play a central role to solve different problems in computer vision. A particular important class of such methods rely on graph models where an object is decomposed into a number of parts, each one being represented by a graph vertex. A graph model-based tracking algorithm has been recently introduced in which a model is generated for a given frame (reference frame) and used to track a target object in the subsequent ones. Because the view of an object changes along the video sequence, the solution updated the model using affine transformations. This paper proposes a different approach and improves the previous one in several ways. Firstly, instead of updating the model, each analyzed frame is backmapped to the model space, thus providing more robustness to the method because model parameters do not have to be modified. A different method for model generation based on user traces has also been implemented and used. This model generation approach is much simpler and user-friendly. Finally, a graph-matching algorithm that has been recently proposed is used for object tracking. This new algorithm is more efficient and leads to better matching results. Experimental results using synthetic and real sequences from the CAVIAR project are shown and discussed. .</abstract>
		<usergroup>hirata@ime.usp.br administrator</usergroup>