Sistem za razpoznavo igralnih kart na osnovi globokega učenja

zaključna naloga Univerzitetnega študijskega programa I. stopnje Strojništvo - Razvojno raziskovalni program

Matej Čampelj (Author), Rok Vrabič (Mentor)

Abstract

Osrednji problem avtomatizacije namiznih iger s kartami je prepoznavanje kart, ki je kompleksna operacija in potrebuje veliko učno množico. Zaradi izboljšav GPU računalnikov lahko te izdelujemo sintetično z uporabo knjižnic, kot je OpenCV. Osredotočili smo se na prepoznavnje kart igre Tarok z namenom avtomatizacije štetja točk po partiji. Napisali smo program, ki prepoznava poljubno število kart preko kamere, shranjuje njihova imena, sešteva njihove točke in preverja bonuse. Uporabili smo detekcijski algoritem YOLO, naučen na umetni množici. Njegova natančnost je bila preverjena in izboljšana s pomočjo realnih slik.

Keywords

diplomske naloge;globoko učenje;slikovni sistemi;prepoznavanje objektov;Tarok;YOLO;OpenCV;

Data

Language:	Slovenian
Year of publishing:	2022
Typology:	2.11 - Undergraduate Thesis
Organization:	UL FS - Faculty of Mechanical Engineering
Publisher:	[M. Čampelj]
UDC:	004.932:004.85(043.2)
COBISS:	133565699
Views:	105
Downloads:	36
Average score:	0 (0 votes)
Metadata:

Other data

Secondary language:	English
Secondary title:	Deep learning-based playing card recognition system
Secondary abstract:	The main problem of the automation of board games with cards is card recognition, which is a complex operation and needs a large learning dataset. Due to improvements in computer GPUs, these can be generated synthetically using libraries, such as OpenCV. We focused on the recognition of the cards of the Tarock card game with the aim of automating the counting of points after the game. We wrote a program that recognizes any number of cards through the camera, stores their names, adds up their points and checks for bonuses. We used the YOLO detection algorithm, which has been trained on an artificial dataset. Its accuracy has been checked and improved upon using real pictures.
Secondary keywords:	thesis;deep learning;imaging systems;object recognition;OpenCV;
Type (COBISS):	Final paper
Study programme:	0
Embargo end date (OpenAIRE):	1970-01-01
Thesis comment:	Univ. v Ljubljani, Fak. za strojništvo
Pages:	VII, 23 f.
ID:	16420300