Prepoznavanje prometnih znakova primjenom računalnog vida

master's thesis

• Title (english): Traffic Signs Recognition Using Computer Vision
• Author: Filip Antonio Pavlinić
• Mentor: Edouard Ivanjko (mentor)
• Committee member: Mario Muštra (predsjednik povjerenstva)
• Committee member: Marko Ševrović (član povjerenstva)
• Committee member: Tonči Carić (član povjerenstva)
• Granter: University of Zagreb; Faculty of Transport and Traffic Sciences; (Division of Intelligent Transport Systems and Logistics); (Department of Intelligent Transport Systems); Zagreb
• Defense date and country: 2021-09-22, Croatia
• Scientific / art field, discipline and subdiscipline: TECHNICAL SCIENCES; Traffic and Transport Technology; Intelligent Transport Systems and Logistics

Abstract

Napredni sustavi za pomoć vozaču neizostavan su dio suvremenih vozila. Sustav za prepoznavanje prometnih znakova omogućuje razvoj autonomnih vozila i smanjenje broja prometnih nesreća uzrokovanih nepoštivanjem prometnih pravila. Prometni znakovi razlikuju se prema boji i obliku znaka te simbolima na samome znaku. Za prepoznavanje prometnih znakova koriste se metode iz područja umjetne inteligencije, odnosno računalnog vida. Za pripremu podataka za učenje modela koriste se metode predobrade podatak i metode za proširenje skupa podataka. Sustav za prepoznavanje sastoji se od modela za otkrivanje i modela za klasifikaciju prometnih znakova. YOLO algoritam korišten je za otkrivanje prometnih znakova i predviđanje njihove lokacije. Za klasifikaciju prometnih znakova u 43 kategorije korištene su konvolucijske neuronske mreže. Prepoznavanje prometnih znakova vrši se na stvarnim video snimkama, a sustav prepoznavanja evaluiran je pri različitim brzinama vozila, različitim intenzitetima svjetlosti, na različitim prometnim znakovima u različitim uvjetima očuvanosti i pri različitim položajima promatranja. Evaluacija je pokazala visoku točnost u radu sustava, no i ukazala na probleme rada u vanjskoj okolini i mogućnosti unaprjeđenja.

Abstract (english)

Advanced driver assistance systems are an indispensable part of modern vehicles. The traffic sign recognition system enables the development of autonomous vehicles and the reduction of the number of traffic accidents caused by non-compliance with traffic rules. Traffic signs differ according to the color and shape of the sign and the symbols on the sign itself. Methods in the field of artificial intelligence, i.e., computer vision, are used to recognize traffic signs. Preprocessing and augmentation methods are used to prepare data for learning the model. The recognition system consists of two models, one for detecting and one for classifying traffic signs. The YOLO algorithm was used to detect traffic signs and predict their location. Convolutional neural networks were used to classify traffic signs into 43 categories from cut out bounding boxes. Traffic sign recognition is performed on real videos, and the recognition system was evaluated at different vehicle speeds, different light intensities, on different traffic signs in different preservation conditions and from different observation positions. The evaluation showed high accuracy in the operation of the system, but also pointed out problems working in outdoor environments and opportunities for improvement.

• Keywords: napredni sustavi za pomoć vozaču; računalni vid; prometni znakovi; konvolucijske neuronske mreže; YOLO algoritam
• Keywords (english): Advanced Driver Assistance Systems; Computer Vision; Traffic Signs; Convolutional Neural Networks; YOLO algorithm
• Language: croatian
• URN:NBN: urn:nbn:hr:119:235925
• Study programme: Title: Intelligent Transport Systems and Logistics; Study programme type: university; Study level: graduate; Academic / professional title: magistar/magistra inženjer/inženjerka prometa (magistar/magistra inženjer/inženjerka prometa)
• Type of resource: Text
• File origin: Born digital
• Access conditions: Open access
• Created on: 2021-10-25 15:11:52