Primerjava različnih načinov učenja globokih nevronskih mrež v avtonomni vožnji

magistrsko delo

Andraž Skupek (Author), Aleš Holobar (Mentor)

Abstract

V magistrskem delu opisujemo avtonomno vožnjo, algoritme za učenje avtonomnih vozil ter algoritme za razpoznavo prometnih znakov. Za implementacijo smo uporabili dva različna načina učenja avtonomnih vozil, in sicer posnemajoče učenje – za implementacijo katerega smo uporabili konvolucijske nevronske mreže, ter samoojačitveno učenje, kjer uporabljamo nevronsko mrežo, model pa se uči iz lastnih napak. Ob implementaciji avtonomnih vozil smo s pomočjo konvolucijskih nevronskih mrež implementirali tudi modele za razpoznavo prometnih znakov. Omenjene modele smo nato združili z algoritmi avtonomne vožnje in s tem dobili vozilo, ki se je sposobno v simulatorju samostojno premikati ter pospeševati ali zavirati glede na razpoznani prometni znak. Modele obeh načinov avtonomne vožnje testiramo na osmih različnih progah, kjer hitrost vožnje upravljamo tudi s pomočjo razpoznavalnika prometnih znakov. Modeli so uspešni, če uspešno prevozijo celotno progo. Rezultati naših modelov so uspešni, saj je kar nekaj modelov uspešno premagalo vseh osem prog.

Keywords

avtonomna vožnja;globoko učenje;nevronske mreže;konvolucijske nevronske mreže;magistrske naloge;

Data

Language:	Slovenian
Year of publishing:	2022
Typology:	2.09 - Master's Thesis
Organization:	UM FERI - Faculty of Electrical Engineering and Computer Science
Publisher:	[A. Skupek]
UDC:	004.85:004.032.26(043.2)
COBISS:	103252483
Views:	107
Downloads:	24
Average score:	0 (0 votes)
Metadata:

Other data

Secondary language:	English
Secondary title:	Comparison of different deep neural network learning algorithms in autonomous driving
Secondary abstract:	The goal of this thesis was to study and demonstrate the performance of algorithms for autonomous driving and algorithms for traffic signs detection and recognition. We used two approaches to autonomous driving. The first one is called behavioural cloning and is using convolutional neural networks. The second one is reinforcement learning. We also implemented models for traffic signs detection and recognition. Finally, we combined those models with autonomous driving models and we simulated the control of an autonomous car which can accelerate and brake according to a recognized traffic signs. All autonomous driving models were tested on eight different simulated tracks, on which the speed of driving was controlled by traffic sign detection models. The model was marked as successful when it successfully completed the track. As demonstrated by our results, several models successfully completed all eight test tracks.
Secondary keywords:	autonomus driving;neural networks;convolutional neural networks;
Type (COBISS):	Master's thesis/paper
Thesis comment:	Univ. v Mariboru, Fak. za elektrotehniko, računalništvo in informatiko, Računalništvo in informacijske tehnologije
Pages:	1 spletni vir (1 datoteka PDF (X, 75 f.))
ID:	14575160