Avtomatsko nastavljanje parametrov segmentacijske metode aplikacije virtualna tipkovnica s pomočjo nevronske mreže

Javornik, Aljaž

| | SLO | ENG | Cookies and privacy

Bigger font | Smaller font

First page > Show document

Show document

Title:	Avtomatsko nastavljanje parametrov segmentacijske metode aplikacije virtualna tipkovnica s pomočjo nevronske mreže
Authors:	ID Javornik, Aljaž (Author) ID Potočnik, Božidar (Mentor) More about this mentor... ID Šavc, Martin (Comentor)
Files:	UN_Javornik_Aljaz_2017.pdf (2,65 MB) MD5: 46765DD4BB55637401B76AC74E136B7A
Language:	Slovenian
Work type:	Bachelor thesis/paper
Typology:	2.11 - Undergraduate Thesis
Organization:	FERI - Faculty of Electrical Engineering and Computer Science
Abstract:	V tem diplomskem delu smo nadgradili izhodiščno aplikacijo Virtualna tipkovnica, ki je namenjena zaznavanju pritiska tipke v nadzorovanih okoliščinah s pomočjo interakcijske površine in vanjo usmerjene kamere. Preučili in implementirali smo metode, ki omogočajo avtomatizirano nastavljanje parametrov segmentacijske metode za izhodiščno aplikacijo. S tem smo izboljšali njeno delovanje v spremenljivih osvetlitvenih razmerah ter ob uporabi drugačnih ozadij interakcijske površine. Preučili in implementirali smo tudi tehnike predobdelave slik ter možnosti segmentacije s pomočjo konvolucijskih nevronskih mrež. Razvili in implementirali smo naslednje rešitve: i) model RGB smo nadomestili z barvnim modelom HSV, ii) rešitev z uporabo metode fotometrične normalizacije, imenovane eno-nivojski retineks, ter iii) rešitev, kjer segmentacijo izvedemo s pomočjo globoke nevronske mreže. Vse nadgradnje osnovne aplikacije Virtualna tipkovnica smo validirali z množico eksperimentov, v katerih smo variirali osvetlitev, oddaljenost svetila, ozadje interakcijske površine ter uporabnika. Kot najuspešnejši nadgradnji sta se izkazali uporaba konvolucijskih nevronskih mrež in morfološka operacija odpiranja.
Keywords:	virtualna tipkovnica, segmentacija slik, nevronska mreža, globoka nevronska mreža, konvolucijska nevronska mreža, kompenzacija osvetlitve, predobdelava, eno-nivojski retineks
Place of publishing:	[Maribor
Publisher:	A. Javornik
Year of publishing:	2017
PID:	20.500.12556/DKUM-65053
UDC:	004.353.4:004.946.5(043.2)
COBISS.SI-ID:	20435478
NUK URN:	URN:SI:UM:DK:FXOAGODP
Publication date in DKUM:	28.02.2017
Views:	1820
Downloads:	189
Metadata:
Categories:	KTFMB - FERI
:	Copy citation

Average score:	(0 votes)
Your score:	Voting is allowed only for logged in users.
Share:

Hover the mouse pointer over a document title to show the abstract or click on the title to get all document metadata.

Secondary language

Language:	English
Title:	Automatic parameters adjustment in segmentation method of virtual keyboard application by using neural network
Abstract:	In this diploma thesis we upgraded the baseline Virtual keyboard application, which is intended to detect key press with a camera pointed onto an interaction surface in controlled conditions. We studied and implemented methods that enable automatic parameters adjustment in segmentation method of our baseline application. These methods lead to better results in variable lightning conditions or in different interaction surface background usage. We studied and implemented methods of image preprocessing and possibilities of image segmentation with assistance of a convolutional neural network. We developed and implemented the following solutions: i) a solution where we replaced the RGB color model with HSV color model, ii) a solution where we used photometric normalization method named Single Scale Retinex, and iii) a solution where we segmented an image with the use of a deep neural network. We validated all the upgrades of our baseline application with a group of experiments in which we varied lightning, light source distance, interaction surface background and user. Convolutional neural network usage and opening morphological operation turned out as the most successful upgrades.
Keywords:	virtual keyboard, image segmentation, neural network, deep neural network, convolutional neural network, illumination compensation, preprocessing, single scale retinex

Comments

Leave comment

You must log in to leave a comment.

Comments (0)

0 - 0 / 0

There are no comments!

Back