Detekcija i raspoznavanje prometnih znakova umjetnim neuronskim mrežama

scientific master's theses

• Title (english): Traffic sign detection and recognition using neural networks
• Author: Tomislav Fištrek
• Mentor: Sven Lončarić (mentor)
• Committee member: Sven Lončarić (predsjednik povjerenstva)
• Granter: University of Zagreb; Faculty of Electrical Engineering and Computing; (Department of Electronic Systems and Information Processing); Zagreb
• Defense date and country: 2011, Croatia
• Scientific / art field, discipline and subdiscipline: TECHNICAL SCIENCES; Electrical Engineering
• Universal decimal classification (UDC): 621.3 - Electrical engineering

Abstract

Rješenje raspoznavanja prometnih znakova ograničavanja brzine realizirano je u dva osnovna dijela. Prvi dio je detekcija ruba znaka koja uključuje segmentaciju velikim skupom različitih pragova u različitim prostorima boja. Za svaku ulaznu sliku pogodan prag i komponenta iz nekog od prostora boja odabire se promatranjem histograma cijele slike. Posebno istrenirana umjetna neuronska mreža odlučuje o odabiru. Na taj način dobijamo sustav koji je prilagodljiv različitim uvjetima osvjetljenja te se segmentacija uspješno provodi i u uvjetima noćne rasvjete gdje prevladavaju crveni tonovi. Nagon segmentacije u prvom dijelu provodi se detekcija Houhovom transformacijom za kružnice u većem opsegu raspona radijusa i zbog toga je sustav prilagodljiv različitim veličinama prometnog znaka. Drugi dio je raspoznavanje gdje je najvažniji dio odabir pravih uzoraka i značajki. Prvi dio, tj. modul za detekciju daje jednu ili dvije najbolje ocjenjene kružnice u segmentiranoj slici smanjene rezolucije. Sada se u originalnoj slici nalazi odgovarajući okvir i on se smanjuje na određenu dimenziju te prebacuje u sivu sliku. Prema tako dobivenoj unutrašnjosti prometnog znaka sustav automatski priprema ulaze za NN koja raspoznaje znakove.

Abstract (english)

Speed limit traffic sign detection is realized in two basic parts. The first part is the detection of traffic sign edge which includes the segmentation by means of a large set of different thresholds in different colour spaces. For each input image a convenient threshold and colour component from one of colour spaces is chosen by the observation of the histogram of the whole image. Specially trained artificial neural network decides about the choice. In this way a system that is adaptable to different lighting conditions is obtained so that the segmentation is successfully carried out even in night lighting conditions where red colours prevail. After the segmentation in the first part the detection by means of the circular Hough transformation is carried out in larger radius range and that is why the system is adaptable to the different sizes of the traffic sign. The second part is the recognition where the most important part is the choice of real patterns and features. The first part, i.e. the detection module results in one or two best graded circles in the segmented image of the reduced resolution. Now, the suitable frame is within the original image and it is reduced to a certain dimension and is transfered into the grey image. The system automatically prepares the inputs for NN in accordance with so obtained interior of the traffic sign.

• Keywords: raspoznavanje uzoraka; analiza slike; detekcija prometnih znakova; raspoznavanje; umjetna neuronska mreža (NM); izdvajanje značajki; Hougova kružna transformacija
• Keywords (english): pattern recognition; image analysis; traffic sign detection; recognition; artificial neural network (NN); extracting features; Hough transform for circles
• Language: croatian
• URN:NBN: urn:nbn:hr:168:877810
• Study programme: Title: Postgraduate master programme in electrical engineering; Study programme type: university; Study level: postgraduate; Academic / professional title: Magistar znanosti elektrotehnike (Magistar znanosti elektrotehnike)
• Catalog URL: http://lib.fer.hr/cgi-bin/koha/opac-detail.pl?biblionumber=36749
• Type of resource: Text
• Extent: 68 str ; 30 cm
• File origin: Born digital
• Access conditions: Closed access
• Created on: 2020-08-24 10:49:40