System for traffic sign recognition

Zadatak ovog završnog rada je programski ostvariti sustav za raspoznavanje prometnih znakova u slikama snimljenim iz vozila u pokretu i ispitati ponašanje sustava na javno dostupnim skupovima slika. U završnom radu također se daje uvid u teme umjetnih neuronskih mreža, strojnog i dubokog učenja te konvolucijskih neuronskih mreža. Govori se o arhitekturi neuronskih mreža i o procesu učenja i testiranja istih, o algoritmima strojnog učenja te usporedbi strojnog s dubokim učenjem. Za ostvarenje sustava u programskom jeziku Python korišten je model konvolucijske neuronske mreže, te je stoga dan uvid u arhitekturu konvolucijskih neuronskih mreža. Naposljetku, dani su implementacija i rezultati sustava.The task of this final thesis is to programmatically implement a system for traffic sign recognition in images captured from moving vehicles and examine the behavior of the system on publicly available image datasets. The thesis also provides insights into the topics of artificial neural networks, machine learning, deep learning, and convolutional neural networks. It discusses the architecture of neural networks and the process of training and testing them, as well as machine learning algorithms and a comparison between machine learning and deep learning. To implement the system in the Python programming language, a convolutional neural network model was used, and therefore an overview of convolutional neural network architecture is provided. Finally, the thesis presents the implementation and results of the system