@InProceedings{Havlena-Torii-ECCV-2010,
  IS = { zkontrolovano 31 Jan 2011 },
  UPDATE  = { 2010-09-16 },
  author =      {Havlena, Michal and Torii, Akihiko and Pajdla, Tom{\'a}{\v s}},
  title =       {Efficient Structure from Motion by Graph Optimization},
  c_title =     {Efektivn{\'\i} tvar z pohybu grafovou optimalizac{\'\i}},
  year =        {2010},
  pages =       {100-113},
  booktitle =   {Computer Vision - {ECCV 2010}, 11th European Conference on
                 Computer Vision, Proceedings, Part {II}},
  editor =      {Kostas Daniilidis and Petros Maragos and Nikos Paragios},
  publisher =   {Springer-Verlag},
  address =     {Berlin, Germany},
  isbn =        {978-3-642-15551-2},
  issn =        {0302-9743},
  series =      {Lecture Notes in Computer Science},
  volume =      {6312},
  book_pages =  {813},
  month =       {September},
  day =         {5-11},
  venue =       {Hersonissos, Greece},
  organization ={Foundation for Research and Technology-Hellas (FORTH)},
  annote = {We present an efficient structure from motion algorithm that
    can deal with large image collections in a fraction of time and effort of
    previous approaches while providing comparable quality of the scene and
    camera reconstruction. First, we employ fast image indexing using large
    image vocabularies to measure visual overlap of images without running
    actual image matching. Then, we select a small subset from the set of
    input images by computing its approximate minimal connected dominating
    set by a fast polynomial algorithm. Finally, we use task prioritization
    to avoid spending too much time in a few difficult matching problems
    instead of exploring other easier options. Thus we avoid wasting time on
    image pairs with low chance of success and avoid matching of highly
    redundant images of landmarks. We present results for several challenging
    sets of thousands of perspective as well as omnidirectional images.},
  c_annote = {P{\v r}edstavujeme efektivn{\'\i} algoritmus pro tvar z pohybu,
    kter{\'y} se vypo{\v r}{\'a}d{\'a} s rozs{\'a}hl{\'y}mi obrazov{\'y}mi
    kolekcemi ve zlomku {\v c}asu i {\'u}sil{\'\i} ve srovn{\'a}n{\'\i} s
    p{\v r}edchoz{\'\i}mi p{\v r}{\'\i}stupy p{\v r}i zachov{\'a}n{\'\i}
    kvality rekonstrukce sc{\'e}ny i poloh kamer. Nejprve u{\v z}ijeme
    rychlou obrazovou indexaci s velk{\'y}m obrazov{\'y}m slovn{\'\i}kem ke
    zm{\v e}{\v r}en{\'\i} vizu{\'a}ln{\'\i}ho  p{\v r}ekryvu obraz{\r u}
    bez {\v r}e{\v s}en{\'\i} samotn{\'e}ho probl{\'e}mu korespondence.
    Potom vybereme malou podmno{\v z}inu mno{\v z}iny vstupn{\'\i}ch
    obraz{\r u} v{\'y}po{\v c}tem jej{\'\i} p{\v r}ibli{\v z}n{\'e}
    minim{\'a}ln{\'\i} souvisl{\'e} dominantn{\'\i} mno{\v z}iny
    rychl{\'y}m polynomi{\'a}ln{\'\i}m algoritmem. Nakonec pou{\v z}ijeme
    uspo{\v r}{\'a}d{\'a}n{\'\i} {\'u}kol{\r u} podle priorit, abychom
    nestr{\'a}vili p{\v r}{\'\i}li{\v s} mnoho {\v c}asu na n{\v e}kolika
    m{\'a}lo t{\v e}{\v z}k{\' y}ch probl{\'e}mech m{\'\i}sto
    zkoum{\'a}n{\'\i} jin{\'y}ch snadn{\v e}j{\v s}{\'\i}ch cest. Vyhmene
    se tedy pl{\'y}tv{\'a}n{\'\i} {\v c}asem na p{\'a}rech obraz{\r u} s
    n{\'\i}zkou {\v s}anc{\'\i} na {\'u}sp{\v e}ch a tak{\'e}
    p{\'a}rov{\'a}n{\'\i} vysoce redundantn{\'\i}ch obraz{\r u} pam{\'a}tek.
    P{\v r}in{\'a}{\v s}{\'\i}me v{\'y}sledky na n{\v e}kolika
    obt{\'\i}{\v z}n{\'y}ch mno{\v z}in{\'a}ch tis{\'\i}c{\r u}
    perspektivn{\'\i}ch i v{\v s}esm{\v e}rov{\'y}ch obraz{\r u}.},
  keywords =    {Structure from motion, Image set reduction, 
                 Task prioritization, Omnidirectional vision},
  prestige =    {important},
  authorship =  {34-33-33},
  project =     {FP7-ICT-247525 HUMAVIPS, MSM6840770038},
  psurl =       {[10.1007/978-3-642-15552-9.pdf]},
}