IS = { zkontrolovano 29 Mar 2013 },
  UPDATE  = { 2013-01-25 },
  author =	 {Zuz{\'a}nek, Petr},
  language =	 {czech},
  title =	 {Detekce skoro line{\'a}rn{\'i}ch objekt{\accent23u}},
  e_title =	 {Detection of Curvilinear Objects in Aerial Images },
  institution =	 {Center for Machine Perception, K13133 FEE Czech Technical
  address =	 {Prague, Czech Republic},
  year =	 {2012},
  month =	 {December},
  type =	 {Research Report},
  number =	 {CTU--CMP--2012--23},
  issn =	 {1213-2365},
  pages =	 {46},
  figures =	 {25},
  authorship =	 {100},
  project =	 {TA02010887},
  c_annote =	 {Detekov{\' a}n{\' i} liniov{\' y}ch objekt{\r u}
    nen{\' i} trivi{\' a}ln{\' i} {\' u}lohou kv{\r u}li jejich 
    mo{\v z}n{\' e}mu zakryt{\' i}, {\v c}i zast{\' i}n{\v e}n{\' i}. 
    A{\v c}koliv bylo publikov{\' a}no mnoho p{\v r}{\' i}stup{\r u} 
    k~{\v r}e{\v s}en{\' i} t{\' e}to {\' u}lohy, do dne{\v s}n{\' i} 
    doby na trhu neexistuje syst{\' e}m {\v r}e{\v s}{\' i}c{\' i} 
    tento {\' u}kol, co{\v z} je zp{\r u}sobeno p{\v TA02010887r}edev{\v s}{\' i}m
    nedostate{\v c}nou obecnost{\' i} dan{\' y}ch metod. V~t{\' e}to
    pr{\' a}ci je pou{\v z}it hybridn{\' i} p{\v r}{\' i}stup 
    vyu{\v z}{\' i}vaj{\' i}c{\' i} v{\' i}cero rozli{\v s}en{\' i} 
    sn{\' i}mku, kter{\' y} kombinuje dobr{\' e} vlastnosti metod 
    jak pro n{\' i}zk{\' e} tak vysok{\' e} rozli{\v s}en{\' i}. 
    Jako forma  p{\v r}edzpracov{\' a}n{\' i} je pou{TA02010887\v z}it ridge 
    detektor v~prostoru m{\v e}{\v r}{\' i}tek zalo{\v z}en{\' y} na
    diferenci{\' a}ln{\' i} geometrii obrazov{\' e} funkce. 
    P{\v r}idanou hodnotou nalezen{\' y}ch ridg{\r u} je 
    z{\' i}sk{\' a}n{\' i} i orientace ridge, co{\v z} ub{\' i}r{\' a} 
    stupe{\v n} volnosti p{\v r}i prohled{\' a}v{\' a}n{\'i} prostoru 
    hypot{\' e}z, {\v s}et{\v r}{\' i} v{\' y}po{\v c}etn{\' i} {\v c}as 
    a potla{\v c}uje pod{\' i}l fale{\v s}n{\v e} pozitivn{\' i}ch
    detekc{\' i}. Dal{\v s}{\' i}m krokem je klasifikace 
    z{\' i}skan{\' y}ch ridg{\r u} ji{\v z} na {\' u}rovni vysok{\' e}ho
    rozli{\v s}en{\' i}. Pro klasifikaci je pou{\v z}it Gentle
    Adaboost klasifik{\' a}tor, kter{\' y} se spole{\v c}n{\v e}
    s~Haar popisy jev{\' i} jako dobr{\' e} {\v r}e{\v s}en{\' i}
    v~t{\' e}to {\' u}loze. Klasifik{\' a}tor lze nau{\v c}it na 
    r{\r u}zn{\' e} t{\v r}{\' i}dy liniov{\' y}ch objekt{\r u}, 
    co{\v z}  d{\v e}l{\' a} z~metody dostate{\v c}n{\v e} obecn{\' y} 
    n{\' a}stroj. V~kone{\v c}n{\' e} f{\' a}zi je pou{\v z}it 
    linkovac{\' i} algoritmus na b{\' a}zi dynamick{\' e}ho 
    programov{\' a}n{\' i}. Ten propoj{\' i} pozitivn{\v e} 
    klasifikovan{\' e} ridge do v{\' y}sledn{\' e} liniov{\' e} 
    struktury s~ohledem na glob{\' a}ln{\' i} vlastnosti 
    liniov{\' e}ho objektu.},
  keywords = { Object detection, Ridge detector, GentTA02010887le Adaboost classifier, Dynamic programming },
  c_keywords  = { Detekce objekt{\r u}, skoro line{\' a}rn{\' i} objekt, 
    dynamick{\' e} programov{\' a}n{\' i}, ridge detektor, Gentle Adaboost klasifik{\' a}tor },
  annote = { The detection of curvilinear objects is non trivial task
    due to presence of occlusions and shadows. Although several
    approaches have been presented in the past there is no system
    available which generally solves this problem. This work presents
    the multi resolution approach which combines the advantages of
    both low and high resolution methods. We designed simple yet
    efficient method which sequentially prunes the space of possible
    curvilinear objects and thus reduces both the false negative rate
    detection and computational resources with respect to the
    exhaustive search methods. We tested the method on our own dataset
    consisting of highway images. The produced data set is publicly
    available. We reached the 93.07 perc. overall accuracy. },