DETERMINATION OF CIRCULAR TRAFFIC SIGNS POSITION BY USING COMPUTER VISION PROCEDURES

V diplomskem delu se ukvarjamo z lokaliziranjem okroglih cestno-prometnih znakov s postopki računalniškega vida. Najprej pregledamo obstoječe rešitve lokaliziranja prometnih znakov in iskanja krožnic ter elips v digitalnih slikah. V nadaljevanju opišemo program, ki smo ga razvili za določanje položaja okroglih cestno-prometnih znakov. Najpomembnejši del programa je iskanje okroglih oblik, ki smo ga izvedli s Houghovo transformacijo za iskanje krožnic in elips. V diplomskem delu so podrobno opisani vsi deli programa, kot so predobdelava slike, segmentacija s Houghovo transformacijo in vizualizacija rezultatov. Program smo preizkusili na 118 slikah, ki vsebujejo enega ali več različno ohranjenih prometnih znakov, slikanih v različnih vremenskih pogojih. Uspešnost lokaliziranja okroglih prometnih znakov je okrog 82 %, pri čemer so v tej statistiki zajete vse slike. Na osnovi dobljenih rezultatov smo predstavili še prednosti, slabosti in možne izboljšave našega pristopa.We are dealing with determination of circular traffic signs position by using the computer vision procedures in this diploma thesis. First, we review existing solutions for traffic signs localization and solutions for searching circles and ellipses in digital images. In the sequel, we describe software that was developed for determination of circular traffic signs position. The most important part of this program is a searching of circular shapes that was implemented by using Hough transform for detecting circles and ellipses. This diploma thesis describes in detail all parts of the program, such as image preprocessing, segmentation by using Hough transform, and visualization of results. This software was tested on 118 images, containing one or more differently preserved traffic signs, filmed in different weather conditions. An efficiency of circular traffic signs localization is around 82%, whereat all images are included in this statistics. Based on obtained results we have presented also the strengths, weaknesses, and possible improvements of our approach