Klasifikacija prometnih znakova korištenjem konvolucijskih neuronskih mreža

Matoš, Ivona

master's thesis

Klasifikacija prometnih znakova korištenjem konvolucijskih neuronskih mreža

Osijek: Josip Juraj Strossmayer University of Osijek, Faculty of Electrical Engineering, Computer Science and Information Technology Osijek, 2020. urn:nbn:hr:200:723809

Matoš, Ivona

Josip Juraj Strossmayer University of Osijek
Faculty of Electrical Engineering, Computer Science and Information Technology Osijek
Department of Software Engineering
Chair of Programming Languages and Systems

Cite this document

APA 6th Edition

Matoš, I. (2020). Klasifikacija prometnih znakova korištenjem konvolucijskih neuronskih mreža (Master's thesis). Osijek: Josip Juraj Strossmayer University of Osijek, Faculty of Electrical Engineering, Computer Science and Information Technology Osijek. Retrieved from https://urn.nsk.hr/urn:nbn:hr:200:723809

MLA 8th Edition

Matoš, Ivona. "Klasifikacija prometnih znakova korištenjem konvolucijskih neuronskih mreža." Master's thesis, Josip Juraj Strossmayer University of Osijek, Faculty of Electrical Engineering, Computer Science and Information Technology Osijek, 2020. https://urn.nsk.hr/urn:nbn:hr:200:723809

Chicago 17th Edition

Harvard

Matoš, I. (2020). 'Klasifikacija prometnih znakova korištenjem konvolucijskih neuronskih mreža', Master's thesis, Josip Juraj Strossmayer University of Osijek, Faculty of Electrical Engineering, Computer Science and Information Technology Osijek, accessed 21 May 2024, https://urn.nsk.hr/urn:nbn:hr:200:723809

Vancouver

Matoš I. Klasifikacija prometnih znakova korištenjem konvolucijskih neuronskih mreža [Master's thesis]. Osijek: Josip Juraj Strossmayer University of Osijek, Faculty of Electrical Engineering, Computer Science and Information Technology Osijek; 2020 [cited 2024 May 21] Available at: https://urn.nsk.hr/urn:nbn:hr:200:723809

IEEE

I. Matoš, "Klasifikacija prometnih znakova korištenjem konvolucijskih neuronskih mreža", Master's thesis, Josip Juraj Strossmayer University of Osijek, Faculty of Electrical Engineering, Computer Science and Information Technology Osijek, Osijek, 2020. Available at: https://urn.nsk.hr/urn:nbn:hr:200:723809

Cite this item: https://urn.nsk.hr/urn:nbn:hr:200:723809

Please login to the repository to save this object to your list.

Metadata

Title	Klasifikacija prometnih znakova korištenjem konvolucijskih neuronskih mreža
Title (english)	Traffic sign classification using convolutional neural networks
Author	Ivona Matoš
Mentor	Zdravko Krpić (mentor)
Committee member	Zdravko Krpić (predsjednik povjerenstva)
Committee member	Ratko Grbić (član povjerenstva)
Committee member	Bruno Zorić (član povjerenstva)
Granter	Josip Juraj Strossmayer University of Osijek Faculty of Electrical Engineering, Computer Science and Information Technology Osijek (Department of Software Engineering) (Chair of Programming Languages and Systems) Osijek
Defense date and country	2020-09-30, Croatia
Scientific / art field, discipline and subdiscipline	TECHNICAL SCIENCES Computing Data Processing
Abstract	Prepoznavanje prometnih znakova jedna je od ključnih odrednica potrebnih za realizaciju autonomnih vozila. U ovom radu evaluiran je rad konvolucijske neuronske mreže na problemu lokalizacije i klasifikacije prometnih znakova korištenjem prilagođene baze podataka. Kako bi se navedeni problem uspješno riješio, implementirana je RetinaNat konvolucijska neuronska mreža. Prometni znakovi koji se trebaju prepoznati podijeljeni su u 33 klase, a za automatsku augmentaciju slika i prilagodbu graničnih okvira unutar istih korištena je ImgAug biblioteka. Model RetinaNet mreže implementiran je u python skriptnom jeziku korištenjem Keras API-ja. Algoritam je testiran na posebnoj bazi slika te se, uzimajući u obzir složenost baze podataka, utvrdilo kako predloženi model mreže dobro prepoznaje prometne znakove postižući mAp u iznosu 71,70 %. Pritom ističu se prometni znakovi poput obaveznog zaustavljanja koji postižu gotovo 90 % mAp. Nasuprot tome, pojavljuju se prometni znakovi sa značajno nižom mAp. Uspješnost prepoznavanja prometnih znakova potencijalno bi se mogla poboljšati dodatnim balansiranjem pojedinih klasa baze podataka te korištenjem dubljih konvolucijskih neuronskih mreža kao okosnica za RetinaNet model.
Abstract (english)	Traffic sign recognition is one of the key joints needed for realization of autonomous driving. This master's thesis evaluates the problems of localization and classification of traffic signs using convolutional neural network and custom made traffic sign database. As a part of this thesis, a RetinaNet convolutional neural network was implemented. There are 33 classes of traffic signs that need to be recognized. Also, an automatic augmentation was performed over the database using ImgAug library, whereas RetinaNet model was implemented using Python and Keras API. The algorithm was tested on separated test database. Considering complexity of the created database, it was established, that in most cases traffic signs are correctly recognized. The overall mAP of 71,70 % was achieved. There were traffic sign classes, such as stop, that achieved mAp of almost 90 %. On the contrary, there were classes that achieved signifficatly low mAp. However, the used model could reach better performance by balancing data in several classes of existing database with new images, as well as, using deeper backbones.
Keywords
Keywords (english)
Language	croatian
URN:NBN	urn:nbn:hr:200:723809
Study programme	Title: UNIVERSITY STUDY AUTOMOBILE COMPUTER AND COMMUNICATION Study programme type: university Study level: graduate Academic / professional title: magistar/magistra inženjer/inženjerka automobilskog računarstva i komunikacija (magistar/magistra inženjer/inženjerka automobilskog računarstva i komunikacija)
Type of resource	Text
File origin	Born digital
Access conditions	Open access
Terms of use
Repository	Faculty of Electrical Engineering, Computer Science and Information Technology Osijek
Created on	2020-11-30 14:30:33