Search

Most commercially successful face recognition systems combine information from multiple sensors (2D and 3D, visible light and infrared, etc.) to achieve reliable recognition in various environments. When only a single sensor is available, the robustness as well as efficacy of the recognition process suffer. In this paper, we focus on face recognition using images captured by a single 3D sensor and propose a method based on the use of region covariance matrixes and Gaussian mixture models (GMMs). All steps of the proposed framework are automated, and no metadata, such as pre-annotated eye, nose, or mouth positions is required, while only a very simple clustering-based face detection is performed. The framework computes a set of region covariance descriptors from local regions of different face image representations and then uses the unscented transform to derive low-dimensional feature vectors, which are finally modeled by GMMs. In the last step, a support vector machine classification scheme is used to make a decision about the identity of the input 3D facial image. The proposed framework has several desirable characteristics, such as an inherent mechanism for data fusion/integration (through the region covariance matrixes), the ability to explore facial images at different levels of locality, and the ability to integrate a domain-specific prior knowledge into the modeling procedure. Several normalization techniques are incorporated into the proposed framework to further improve performance. Extensive experiments are performed on three prominent databases (FRGC v2, CASIA, and UMB-DB) yielding competitive results.

V slovenskem jezikovnem prostoru se je sredi 90. let pojavila težnja po enotnem računalniškem zapisu slovenskega govora. Ta bi standardiziral in poenotil raziskovalno delo na področju govornih in jezikovnih tehnologij, ki se nanaša na fonetične označitve slovenskih govorjenih besedil. Sodelovanje mariborske in ljubljanske skupine raziskovalcev (Fakulteta za elektrotehniko, računalništvo in informatiko, Pedagoška fakulteta Fakulteta za elektotehniko, ZRC SAZU, Inštitut za slovenski jezik Frana Ramovša) je rodilo računalniški fonetični zapis simbolov IPA in MRPA. Predvsem slednji so velikega pomena za enostavno in prijazno računalniško podobo. In the Slovene-speaking area in the mid-90's there appeared a trend towards a unfied computer transcription of Slovene speech. This transcription would standardize and unify research in the field of speech and language technologies that touches upon the phonetic symbolization of Slovene spoken texts. The cooperation of the research groups from Maribor and Ljubljana (Faculty of Electrical Engineering and Computer Science Faculty of Education Faculty of Electrical Engineering: The Scientific Research Centre of SASA, Fran Ramovš Institut of the Slovenian language) resulted in the development of computer-based phonetic transcriptions of the symbols in IPA and MRPA. Largely the latter are of great importance for simple and userfriendly computer usage.

Speech corpora are an essential infrastucture element for speech technologies research in a given language. Speech corpora consist of machine-readable speech recordings. They are futher equipped with additional data, such as speaker and recording session descriptors and different segmentation and annotation elements, e.g. graphemic, phonetic and prosodic transcriptions or descriptors. In the article we describe the process of designing a speech corpus - from text or topic selection, recording, segmentation and labelling procedures. We conclude with an overview of speech corpora for the Slovene language. Govorne zbirke so nepogrešljive pri raziskovalnem delu na področju tehnologij govorjenega jezika. Vsebujejo predvsem računalniško berljive posnetke govora. Posnetkom so vedno priloženi še podatki, ki na različne načine opisujejo posneti govor. Priloženi podatki so opisi govornih dejavnikov, dejavnikov govorcev ter zapisi in označitve posnetega govora. Pogosto govorni zbirki priložimo še navodila za uporabo ter rezultate izbranih analiz govornih posnetkov. Članek opisuje osnovne gradnike govornih zbirk ter postopek zasnove, snemanja, segmentacije, označevanja in analize govornih zbirk. V sklepnem delu se nahaja pregled govornih zbirk za slovenski jezik.

In the Slovene-speaking area in the mid-90's there appeared a trend towards a unfied computer transcription of Slovene speech. This transcription would standardize and unify research in the field of speech and language technologies that touches upon the phonetic symbolization of Slovene spoken texts. The cooperation of the research groups from Maribor and Ljubljana (Faculty of Electrical Engineering and Computer Science; Faculty of Education; Faculty of Electrical Engineering: The Scientific Research Centre of SASA, Fran Ramovš Institut of the Slovenian language) resulted in the development of computer-based phonetic transcriptions of the symbols in IPA and MRPA. Largely the latter are of great importance for simple and userfriendly computer usage. V slovenskem jezikovnem prostoru se je sredi 90. let pojavila težnja po enotnem računalniškem zapisu slovenskega govora. Ta bi standardiziral in poenotil raziskovalno delo na področju govornih in jezikovnih tehnologij, ki se nanaša na fonetične označitve slovenskih govorjenih besedil. Sodelovanje mariborske in ljubljanske skupine raziskovalcev (Fakulteta za elektrotehniko, računalništvo in informatiko, Pedagoška fakulteta; Fakulteta za elektotehniko, ZRC SAZU, Inštitut za slovenski jezik Frana Ramovša) je rodilo računalniški fonetični zapis simbolov IPA in MRPA. Predvsem slednji so velikega pomena za enostavno in prijazno računalniško podobo.

V tem diplomskem delu je predstavljena izdelava in delovanje testne naprave za avtomobilsko okrasno luč. Naprava ima 5 funkcij: zatemnjevanje s pulzno-širinsko modulacijo, preverjanje komunikacije z gonilnikom svetlečih diod, testiranje animacije, testiranje zaščite pred pregrevanjem in programiranje nastavitev in naslova gonilniku svetlečih diod. Vse elektronske komponente so morale ustrezati standardom avtomobilske industrije AEC-Q100 in AEC-200. Glavni del naprave je mikrokrmilnik STM Discovery STM32F429I z vgrajenim zaslonom na dotik, preko katerega uporabnik upravlja s funkcijami. Ta komunicira z lučjo preko pretvornika UART - CAN TJA1051. Naprava ima 4-polni vtič in sicer za CAN high, CAN low, termistor in maso. Zapakirana je v robustno ohišje, ki je primerno za uporabo v industrijskem okolju. Program za napravo je bil narejen v programskem okolju CubeIDE. Pri programiranju je bil uporabljen princip realno-časnega operacijskega sistema (angl. Real-time operating system oz. RTOS) za boljšo urejenost in preglednost. Program deluje na principu avtomata stanj, kjer en zaslon grafike predstavlja eno stanje in hkrati eno opravilo realno-časnega operacijskega sistema. Rezultat dela je naprava, ki je zmožna vseh opisanih funkcij. Uporabniku omogoča intuitivno in preprosto uporabo. Merjenje temperature v okviru zaščite pred pregrevanjem ima zadostno natančnost. Uporablja se za testiranje prototipov avtomobilskih luči. This thesis addresses the process of making a testing device of an automotive decorative lamp. The device has 5 functions: dimming with pulse-width modulation, monitoring of a LED driver communication, animation testing, thermal derating test and flashing of settings and device address. All components of the device must comply with automotive standards. Main component of the device is microcontroller STM Discovery STM32F429I with integrated touchscreen, through which user controls the devices functions. The controller communicates with lamp through UART - CAN converter TJA1051. The device has 4-pin connector, a pin for CAN high, pin for CAN low, pin for thermistor and the last pin for ground. Test machine is packed in a robust, industrial grade case. Software was created in CubeIDE program, which is a product of the same manufacturer as the microcontroller and thus already includes plenty of resources for it. Program includes a concept of real time operating system (RTOS) for better structure and clarity of code. Software is written as a state machine, where one screen represents one state and at the same time one task of real time operating system. The result of this thesis is a device, which is capable of all the above-described functions. It is easy to use. Temperature measurement included in thermal derating test is precise enough. The machine is primarily used to test prototypes of car lamps.

V tem dokumentu je opisan postopek izdelave zaključnega dela na prvi stopnji visokošolskega strokovnega študija na fakulteti za elektrotehniko v Ljubljani. Cilj zaključnega dela je razpoznava napak pri predmetih na rentgenskih slikah. Predstavljajmo si, da je pred nami predmet, ki na zunaj izgleda popolnoma nepoškodovan in po predpisih. Pri takem bi si sigurno mislili, da je zmožen iti na prodajno police oziroma v dnevno uporabo. Kaj pa če je predmet od znotraj poškodovan, odkrušen, ipd.? Predmeta ne moremo odpreti, ker je iz enega kosa in ga ne moremo kar odpreti in pogledati vanj, zato tukaj nastane težava. Obstajajo pa tehnološke rešitve, ki so zmožne rešiti ta problem. To je uporaba metod računalniškega in strojnega vida, ki se razvijajo na področju umetne inteligence. Ampak kako bi videli notranjost? To se lahko izvede z zajemom rentgenske slike predmeta. Sprva je bilo potrebno pridobiti slike za podatkovno bazo s katero lahko začnemo učiti stroj. Te slike se je pridobilo z rentgenskim zajemom. Sprva je potrebno osnovne slike programersko izboljšati z različnimi filtri, da je možno iz njih razbrati čim več. Po tem se na slikah izvede maskiranje, saj izvajamo segmentacijo in je potrebno označiti kateri del nas zanima, da se ga stroj nauči napovedovati.. Za programski jezik, ki je bil za ta projekt v uporabi pri programiranju filtrov ter modela nevronskega omrežja, je bil izbran Python, za knjižnico, strojnega učenja pa PyTorch. Model, kateri je bil uporabljen za to delo je U-Net, saj kaže ene izmed najboljših rezultatov za segmentacijo slik. To se je tudi videlo pri rezultatih učenja, ki in kasnejši napovedi z tem modelom nevronskega omrežja. Rezultati napovedi so bili na koncu kar uspešni, vendar zaradi težavnosti vhodnih slik in natančnosti mask je uspešnost malo manjša kot željeno. Izguba se je manjšala z vsako iteracijo pri učenju kar nam je dalo vedeti, da se model U-Net trenira na vhodnih podatkih. Kljub temu, da so vhodne slike težavne, je napovedovalni čas zelo majhen, okoli 0.01s, kar je za uporabo v realnem svetu zelo uporabno. This document describes the process of producing the final work at the first stage of higher vocational studies at the Faculty of Electrical Engineering in Ljubljana. The goal of the final part is to identify errors in objects which are photographed with an X-Ray device. Let's imagine that there is an object in front of us that looks completely unscathed on the outside and by the regulations. I'm sure you'd think it was capable of going on store shelves or day-to-day use. What if the object is damaged from the inside, chipped, etc.? We cannot open an object because it's made out of one piece so we can't just take a look at insides, so there's a problem here. However, there are technological solutions that are capable of solving this problem. This is the done with the help of computer and machine vision methods developed in the field of artificial intelligence. But how would you see the inside? This can be done by captureing an X-ray of the object. At first, it was necessary to obtain images for a database with which we can start learning the machine for later predictions. With the help of my coworkers at INEA d.o.o. we captured these images from an X-Ray at Jožef Štefan Institute. First thing after using filters on these images which help us see it's details better and get better quality images, the masks had to be made, because we will be performing segmentation and it is necessary to indicate which area of the image is in our range of interest so that we can train the machine to predict it. Python was chosen as a programming language, and PyTorch as a library, for machine learning. The model used for this project is U-Net, because it is shown that it has one of the best accuracy of prediction for image segmentation. This was also seen in the results of training process and later shown when predictions were made. The results of predictions were quite successful in the end, but they could have been even better if a larger image database was used to train the model. Moreover due to the difficulty of images the results were not as accurate, but still in acceptable range. Loss was getting lower and lower with each iteration in training which lead us to knowing that the model is successfully training on the given data. The prediction time of around 0.01s is short and implementable in the real world.

Diplomsko delo se ukvarja s posodobitvijo komponente sistema za vodenje železniškega prometa. Ta komponenta je zadolžena za sledenje in opisovanje vlakov v področju vodenja. Pri tem uporablja orodje Drools - mehanizem pravil, ki omogoča učinkovito ločevanje programske kode in logike za delovanje sistema. Mehanizem pravil je ekspertni sistem, ki na podlagi ločeno definiranih pravil in stanja v sistemu izvaja odločitve. Uporaben je predvsem, ko zahteve za delovanje sistema postanejo zelo kompleksne in se pogosto spreminjajo. V tem sistemu je implementiran kot samostojna Java aplikacija s starejšimi različicami programskih orodij. V okviru diplomskega dela je ta aplikacija posodobljena za uporabo novejše različice orodja Drools in orodja Maven za upravljanje s projektom. Tako je pripravljena za izvajanje v novejšem Java okolju. S tem bodo nadaljnje nadgradnje aplikacije enostavnejše, podpora za uporabljena orodja pa bo na voljo dlje. This thesis focuses on upgrading a train traffic control system component, used for object tracking and monitoring of the controlled area. This is achieved using Drools, a rule engine, which can provide efficient separation of execution logic and application code. A rule engine is an expert system that provides decision-making capabilities according to separately defined rules and the current state of the system. This is a better approach for handling complex execution logic that changes frequently. In this system, it is implemented as a standalone Java application. The goal of this thesis is to modify the application to include Maven project management tool and an updated version of Drools. Additionally, the goal is to prepare it for execution in a newer version of the Java runtime environment. This provides a better basis for future upgrades and longer support periods.

Zasnujte in izvedite preizkus izbranih metod samodejnega razpoznavanja varnostno sumljivih zvokov. Za izvedbo metod strojnega učenja izbranih modelov razpoznavalnikov in ugotavljanje njihove natančnosti pridobite zbirko posnetkov varnostno sumljivih zvokov in drugih običajnih zvokov v okolju, ki jo sestavite iz razpoložljivih javno dostopnih tovrstnih posnetkov. Natančnost samodejnega razpoznavanja varnostno sumljivih zvokov ovrednotite z izbranimi merami, ki se uporabljajo na tem področju, ter primerno razdelitvijo uporabljene zbirke zvočnih posnetkov na množico učnih in testnih posnetkov. Design and conduct a test of selected methods of automatic recognition of security-suspicious sounds. To implement machine learning methods of selected recognizer models and determine their accuracy, obtain a collection of recordings of security-suspicious sounds and other common sounds in the environment, which you compile from publicly available recordings of this type. You evaluate the accuracy of the automatic recognition of security-suspicious sounds with the selected measures used in this field, as well as the appropriate division of the used collection of audio recordings into a set of training and test recordings.

V predstavljenem delu smo izvedli, preizkusili in ovrednotili učinkovitost izbranih računskih metod izboljšanja kakovosti govora v zvočnih signalih, pri čemer so bili uporabljeni govorni posnetki v slovenskem jeziku. Metode za izboljšanje kakovosti govora se običajno uporabljajo kot predhodni proces pri sistemih za samodejno razpoznavanje govora, saj se z odpravo motenj in šumov, primešanih govornemu signalu, zmanjša možnost napačnega razpoznavanja govora. Uporaba tovrstnih metod je nepogrešljiva predvsem pri aplikacijah za video klice. V okviru predstavljenega dela smo preizkusili dva modela generativnih nasprotniških nevronskih omrežij, in sicer model nevronskega omrežja SEGAN in model nevronskega omrežja Wave-U-Net. Oba modela smo z metodami strojnega učenja naučili in preizkusili z uporabo slovenske govorne zbirke, ki je bila pridobljena v okviru projekta Razvoj slovenščine v digitalnem okolju (RSDO). Uspešnost uporabljenih modelov in metod smo na koncu ovrednotili in primerjali z merami, ki se običajno uporabljajo za vrednotenje kakovosti zvočnih govornih posnetkov. Analizirali smo delovanje obeh uporabljenih metod in razlike v njuni zmogljivosti pri uporabi govornih posnetkov v slovenskem in angleškem jeziku. In the presented work, we implemented, tested and evaluated the effectiveness of selected computational methods for improving speech quality in audio signals, where we used speech recordings in the Slovenian language. Speech enhancement methods are commonly used as a pre-process in automatic speech recognition systems, as the elimination of disturbances and noises mixed with the speech signal reduces the possibility of incorrect speech recognition. The use of such methods is indispensable, especially in video calling applications. As part of the presented work, we tested two models of generative adversary neural networks, namely the SEGAN neural network model and the Wave-U-Net neural network model. Both models were trained and tested using machine learning methods with the Slovene language speech database, which was acquired as part of the project Development of Slovene in the Digital Environment (RSDO). The performance of the models and methods used was finally evaluated and compared with the measures commonly used to evaluate the quality of speech sound recordings. We analyzed the operation of both methods used and the differences in their performance when using Slovenian and English language speech recordings.

Uporaba tehnologije v šolstvu za namen izobraževanja je skozi čas močno napredovala. Od majhnih glinenih tablic vse do interaktivnih tabel, ki poleg pisanja in brisanja omogočajo tudi multidimenzionalno učenje. Po svoji naravi je pouk socialno dogajanje, v katerem niso pomembne le podane vsebine, temveč tudi medosebni odnosi, ki se v času izvajanja pouka ustvarjajo. Zanj je tako ključnega pomena kakovostna socialna interakcija, ki pa jo učitelj težko ustvari in obdrži, če mora večino svojega časa preživeti s hrbtom obrnjen proti učencem, zaradi stalnega brisanja table. Raziskave so pokazale, da veliko izvajalcev vzgojno-izobraževalnega procesa uporablja samo osnovne funkcije interaktivnih tabel ali pa je zaradi pomanjkanja izobraževanj o njeni pravilni uporabi sploh ne uporabljajo. Prav tako je interaktivna tabla kar precejšen finančni zalogaj za marsikatere izobraževalne ustanove, zato je manj razširjena, kot bi lahko bila. Cilj diplomskega dela je bil oblikovati izboljšave klasične bele table, ki so finančno dostopnejše kot interaktivne table in nudijo izboljšave za vse naštete pomanjkljivosti. Moja pametna tabla učiteljem omogoča kakovostnejše poučevanje, saj skrajša čas, ki ga izvajalec vzgojno-izobraževalnega procesa preživi s hrbtom obrnjen proti učencem, in mu omogoča kakovostnejšo socialno interakcijo z učenci. Zadani cilj sem dosegel z implementacijo mehanizma, zmožnega premikanja v dveh oseh, na katerem sta pritrjena pisalo in goba za brisanje. Za nadzor mehanizma skrbi krmilnik PlanetCNC, ki krmili koračne motorje osi in je povezan na mikro računalnik Raspberry Pi. Upravljanje mehanizma je možno prek aplikacije za mobilni telefon, narejene v programskem okolju MIT App Inventor. Aplikacija omogoča pisanje po tabli, brisanje celotne table in brisanje izbranega dela table. The use of technology in education for educational purposes has advanced greatly over time. From small clay tablets, all the way to interactive whiteboards, which, in addition to writing and erasing, also enable multidimensional learning. By its nature, teaching is a social event in which not only the given content is important, but also the interpersonal relationships that are created during the teaching. For learning, quality social interaction is crucial, but it is difficult for the teacher to create and maintain good interaction if he has to spend most of his time with his back to the students, due to the constant erasing of the board. Research has shown that many teachers use only basic functions of interactive tables, or do not use them at all due to the lack of education on their proper use. Also, the interactive whiteboard is quite a financial burden for many educational institutions, so it is less widespread than it could be. The aim of my thesis is to design improvements of the classic whiteboard, which are more affordable than interactive whiteboards and offer improvements for all the shortcomings listed above. My smartboard enables teachers to teach better, as it shortens time that they spend with their back turned to the students and allows them to have a better social interaction with students. I achieved the set goal by implementing a mechanism capable of moving in two axes, on which a pen and a wiping sponge are attached. The mechanism is controlled by a PlanetCNC controller, which controls the stepper motors of the axis and is connected to a Raspberry Pi microcomputer. The mechanism can be controlled via a mobile phone application made in the MIT App Inventor software environment. The application allows you to write on the board, delete the entire board or delete selected parts of the board.

Izvajanje nadzora nad omrežnim prometom je s pravim orodjem lahko zelo enostavno. Če se uporablja več orodij in več omrežij, ki jih je potrebno nadzorovati, pa lahko tudi z dobrim orodjem pride do težav. Zato se je v Operativnem centru kibernetske varnosti porodila ideja, da bi se razvilo platformo, ki bi vsa nadzorna orodja in nadzorovana omrežja združila v en sistem. S takim orodjem bi se poenostavilo delo nadzornikov in lahko bi se opravljalo povezovanje med različnimi omrežji ter lažje iskalo napake in jih odpravljalo. Zaznavanje in odziv na napake in napade v sistemu bi se s tem skrajšala. Platforma, ki je opisana v zaključni nalogi, bo zbirala podatke iz sistemov SIEM, drugih omrežnih nadzornih naprav, kot so IDS/IPS in požarne pregrade, ter sistemskih strežnikov, kot so DNS, domenski krmilniki in avtentikacijski strežniki. Iz teh podatkov bo platforma delala analize in iskala korelacije med njimi. S tako celovito analizo in pregledom nad omrežjem bo razvidno, kaj se v omrežju dogaja v stvarnem času. Ko bo platforma pripravljena, bo s pomočjo strojnega učenja prepoznala anomalije v omrežju in pri uporabniku. Za znane anomalije bodo postopki njihovega reševanja avtomatizirani, za nove anomalije pa bo avtomatizirano obveščanje pristojnih tehnikov in intervalno stopnjevanje obveščanja. Platforma bo avtomatizirala postopke reševanja različnih anomalij iz različnih omrežij glede na prioritete omrežja. V zaključnem delu je opisano razvijanje platforme, kako se je delo zastavilo ter kaj je namen in cilj platforme. Trenutno je ogrodje platforme že izdelano. Naslednje korake pri razvoju pa narekujejo potrebe, ki se pojavljajo ob razvoju platforme, ter potrebe, ki nastajajo v sami delovni organizaciji. Opisani so tudi različni deli platforme, njihovo delovanje in medsebojno povezovanje. Trenutno je najpomembneje dokončati osnovno funkcionalnost platforme, to je obveščanje o anomalijah. Dodajanje vmesnikov za nadzorne naprave in sisteme SIEM pa bo naslednja stopnja razvoja. Implementing network traffic control can be very easy with the right tools. However, if more tools are used and more networks need to be monitored, problems can also occur with good tools. Therefore, the idea was born in the Cyber Security Operations Center to develop a platform that would combine all monitoring tools and monitored networks into one system. Such a tool would simplify the work of technicians and make it possible to connect between different networks and make it easier to find and correct errors. Detecting and responding to errors and attacks in the system would be shortened. The platform described in the final thesis will collect data from SIEM systems, other network monitoring devices such as IDS / IPS and firewalls, and system servers such as DNS, domain controllers, and authentication servers. From collected data, the platform will do analyses and look for correlations between them. With such a comprehensive analysis and overview of the network, it will be clear what is happening in the network in real time. When the platform is ready, it will use machine learning to find anomalies in the network and users behavior changes. For known anomalies, the procedures for their resolution will be automated, and for new anomalies, the notification of technicians and interval escalation of notification will be automated. The platform will automate the procedures of resolving various anomalies from different networks according to network priorities. The final part describes the development of the platform, how the work was set and what the purpose and goal of the platform is. Currently, the platform framework is already in place. The next steps in development are dictated by the needs that arise during the development of the platform, as well as the needs that arise in the work organization itself. The different parts of the platform, their operation and interconnection are also described. Currently, the most important thing is to complete the basic functionality of the platform, which is to inform about anomalies. Adding interfaces for SIEM control devices and systems will be the next stage of development.

Predmet diplomskega dela je priprava in izvedba projekta pametne hiše na osnovi sistema KNX (standardizirani protokol za kontrolo zgradb). V sklopu projekta pametne hiše so programirana naslednja področja: razsvetljava, senčila in ogrevanje/hlajenje. Varnostni sistem in oddaljeni dostop sta pripravljena v sodelovanju z drugim podjetjem. Celoten projekt je razdeljen na dva dela, in sicer na pripravo v pisarni in implementacijo na terenu. Pri pripravi v pisarni so bili najprej dopolnjeni tlorisi, na podlagi teh je bil v programu ETS5 (program, v katerem se nastavlja, povezuje in programira elemente KNX) izdelan projekt, torej vse povezave med stikali, lučmi, ventili in senčili. Nato so bili na vsak aktuator in druge elemente KNX naloženi podatki o fizičnem naslovu, monterjem je bilo na škatlo označeno, kam je treba elemente montirati. Na koncu sta sledili še dve odpravi na hišo, ki sta obsegali nalaganje programa v sistem, preverjanje delovanja in odpravo napake. Naročnik je pogledal še zadnje podrobnosti in odobril projekt. The subject of the graduation thesis is preparation and implementation of the smart house project, based on KNX system (standarized building control protocol). As part of the smart house project the following areas have been programmed: lighting, blinds and heating/cooling. While security system and remote access has been done in a collaboration with other company. The whole project described is roughly divided into two parts, namely office preparation and field implementation. When preparing in the office, the floor plans had been done first, and a project was build from these in ETS5 (a program in which KNX elements are configured, linked and programmed). So all the connections between switches, lights, valves and blinds had been done in the program. Then the physical address information was uploaded to each actuator and other KNX elements, and the boxes were marked to installers where to mount. Finally, two more expeditions to the house where the program was uploaded into the system followed, and the operation was checked and little errors immediately fixed. The client looked at the latest details and approved the project.

V dobi razvoja in avtomatizacije vsa industrijska podjetja stremijo k boljši učinkovitosti in hitrejši proizvodnji izdelkov. Za realizacijo tega cilja so potrebni dobro programsko orodje in sodobne elektronske naprave, ki to omogočajo. Eden glavnih dejavnikov za avtomatizacijo je upravljanje elektronskih naprav s programsko kodo. Od zagona do uporabe temeljnih funkcionalnosti vsakega elektronskega instrumenta skrbi gonilnik naprave. To je programska oprema, ki z dobro grajenim programskim vmesnikom omogoča komuniciranje in upravljanje naprave z računalnikom. Gonilniki se uporabljajo predvsem pri standardnih napravah, kot so monitor, računalniška miška, tiskalnik, lahko pa jih izvedemo tudi na bolj specifičnih oziroma nestandardnih napravah, med katere uvrščamo laboratorijsko in posebno testno opremo. Izvedba gonilnika mora biti skrbno načrtovana, saj lahko vsaka sprememba v programski kodi kritično vpliva na njegovo delovanje. Je kot steber v zgradbi če je slabo načrtovan, je le vprašanje časa, kdaj se bo zgradba porušila. Vsebuje vsa temeljna opravila, kot so – v primeru močnostnega vira in etalonskega merilnika pogreška – povezovanje, pisanje v napravo in branje iz nje, nastavljanje in tudi prekinitev komunikacije. Na tem se lahko gradijo vse druge funkcije, ki pripomorejo k bolj specifičnim nalogam in zagotavljajo, da delo poteka na daljavo, hitro in brez človekovega posega, saj je to tudi namen avtomatizacije. In the era of development and automation, all industrial companies strive for better efficiency and faster production of their products. For making this goal happen, however, a good and adequate software framework and modern electronic devices are needed. One of the factors that are of quintessence is the management of these instruments with program code. The device driver takes care of booting and all the applications of the basic functionalities. It is a software that, with a well-built software interface, allows users to communicate and manage a distinct device with a computer. Device drivers are mostly used for standard devices such as: monitors, computer mouse, printers, etc., but we can also implement its functionalities on more specific non-standard devices such as laboratory and special test equipment. The implementation of such project must be meticulously and thoughtfully designed, since any further changes in the main-code can cause error in its operation . It is like a pillar in the building, if it is poorly designed, it is just a matter of time before it collapses. The driver contains all basic functions (in this case a power supply and standard error meter) such as: connection, data transfer to the device itself, data receiving, setting values from the devices buffer and also closing its ports and breaking communication. All other functions stem from this concept and contribute to more specific tasks and allow us to communicate with these devices remotely, quickly and without human intervention, as this is also the point of automation.

Lokalizacija na osnovi zvoka postaja vedno bolj razširjena. Kljub temu, da se že uporablja na velikem številu področij, obstaja še veliko možnosti za njeno uporabo. Glavni cilji magistrske naloge so predstavitev postopkov za lokalizacijo oseb v prostoru na osnovi zvokov, ki jih povzročajo, izdelava in ureditev zbirke zvočnih posnetkov ter preveriti nekaj postopkov lokalizacije na ustvarjeni zbirki. V prvem delu so opisani področje in primeri uporabe lokalizacije z zvokom ter različni postopki za njeno izvedbo. Na kratko je predstavljena tudi zgodovina in obstoječa dela na tem področju. V nadaljevanju so podrobneje predstavljeni postopki lokalizacije z uporabo računanja časovne razlike pri prihodu zvoka v posamezen mikrofon (v angleščini so poznani pod kratico TDOA). Predstavljenih je več načinov za računanje časovnega zamika. Predstavljena je tudi geometrija problema in različni načini za izračun položaja govorca. Zadnji del magistrske naloge je namenjen eksperimentalnemu delu. Najprej je predstavljena zbirka zvočnih posnetkov in njeno pridobivanje. V nadaljevanju je opis izvedbe različnih, prej predstavljenih postopkov in preizkus na pridobljeni zbirki podatkov. Sledi še analiza rezultatov in njihovo ovrednotenje. Sound source localization is becoming increasingly popular localization technique that has a wide range of applications. The aim of this master thesis is to present different sound source localization methods by creating and transcribing a collection of audio recordings in order to test some of the described methods. In the first part the main concepts and methods of sound source localization are presented. The examples of the applications of sound localization methods are listed together with the short history overview. The existing work in the field is also described. The thesis gives a more detailed description of the methods based on time difference of arrival in each microphone. The thesis presents different methods for calculating time differences and explains the geometry of the problem along with several methods used to calculate the speaker position. Experimental work is described in the last part of the thesis. First, the sound recordings and protocols used for recording them are presented. Next, the implementation of some previously mentioned methods is described. In conclusion is given the analysis and the evaluation of the results.

Diplomsko delo obravnava avtomatsko razpoznavanje glasov slovenskega jezika na osnovi podatkovne zbirke Sofes. Uporabljen je nabor programskih orodij za razpoznavanje glasov in govora Kaldi, ki do sedaj za slovenski jezik še ni bil uporabljen. Na podlagi različnih jezikovnih in akustičnih modelov razpoznavalnika je bilo izvedeno rapoznavanje glasov. Dobljeni rezultati, tako na podlagi nevronskih omrežij kot klasičnih HMM pristopov, so obetavni. The thesis is focused on automatic speech recognition of Slovenian phonemes, based on the Sofes database. The speech and phoneme recognition toolkit Kaldi is used, which has thus far not been used for the Slovenian language. The speech recognition process was implented with various acousting and lingustic models. The results, obtained by using both neural networks and classical HMM approaches, yielded promising results.

Cilj dela v diplomski nalogi je razviti govorni vmesnik, ki bo uspešno reševal probleme z vnašanjem podatkov v informacijske sisteme med terenskim delom. V prvem delu naloge smo raziskali področje razpoznavanja govora in pregledali možne govorne vmesnike ter orodja, katere bi lahko uporabili pri svojem delu V drugem delu naloge smo se osredotočili na samo izvedbo govornega vmesnika v programskem jeziku Python. Pri obdelavi posnetkov govora smo uporabili nekaj nestandardnih Python knjižnic. Za razpoznavanje govora smo uporabili Googlov govorni programski vmesnik Google Speech API. Razpoznano besedilo smo oblikovali v HTML formatu. Razvili smo tudi grafični vmesnik. Delovanje govornega vmesnika smo preizkusili v okoljih z različno ravnijo hrupa. Ugotovili smo, da zadovoljivo dobro deluje tudi pri posnetkih, narejenih v naravnem okolju, v katerem terensko delo navadno poteka. Main goal of the thesis was to develop a speech interface for solving problems with data entry during fieldwork. In first part of the thesis we did an overview of speech recognition field, tools and speech interfaces which we cloud use in development of my own speech interface. In the second part of the thesis we focused on developing speech interface with python programing language. We used some nonstandard python libraries for audio processing. Speech recognition was performed by Google Speech API. We used HTML format to achieve the desired text structure of the output. We also developed a graphical user interface. We tested the speech interface in different environments with different noise volumes. We concluded that it performs well with voice recordings that were recorded in a natural environment, where fieldwork is usually performed. Performance drops only in environments with a really loud noise.

Avtonomne jadrnice so robotska plovila, ki brez potrebe po človekovih posegih uporabljajo veter za pogon in upravljajo z jadri in krmilom. Namen razvoja teh plovil je izdelava jadrnice brez posadke, ki omogoča popolno samostojno jadranje, varno plovbo in izvajanje dolgoročnih nalog na morju. Najpomembnejši sistemi avtonomnih jadrnic so: sistem za dolgoročno načrtovanje plovne poti v odvisnosti od vremenskih razmer, sistem za kratkoročno preusmerjanje jadrnice v odvisnosti od trenutnih podatkov iz senzorjev, sistem za izogibanje trkom in sistem za upravljanja s plovilom. Uporaba mehke (neizrazite) logike pri programiranju nadzornega sistema za upravljanje aktuatorjev se je izkazala za zelo primerno orodje za prenos ekspertnega znanja izkušenih jadralcev v računalniške programe v obliki preprostih pravil vrste ČE-POTEM. Cilj raziskav, predstavljenih v tem delu, je predlog novih nastavitev jader in aktuatorjev ter predlog sistema upravljanja, ki temelji na mehki logiki. Predlagamo razvoj avtonomne jadrnice s tremi različnimi sprednjimi jedri in možnostjo spreminjanja velikosti glavnega jadra. Možnost uporabe več jader v odvisnosti od vetrovnih pogojev in valov na morju podaljšuje obdobje najbolj učinkovitega delovanja (največje hitrosti) jadranja. Zaradi tega jadrnica z omenjeno postavitvijo jader prikazuje bolj realen model, ki se danes uporablja predvsem za oceanskega potovanja. Simulacije jadrnice s tremi različnimi sprednjimi jedri in možnostjo spreminjanja velikosti glavnega jadra so pokazale, da je uporaba mehke logike v nadzornem sistemu, ki upravlja z aktuatorji, smiselna in odpira možnost za razvoj hitrejšega, bolj stabilnega in energetsko učinkovitega plovila. Autonomous sailing boats are robotic vessels that use the wind to drive and can operate the sails and rudder without the need for human intervention. The purpose of the development of these vessels is constructing boats without crew, enabling full independent sailing, safe navigation and implementation of long-term tasks at sea. The most important systems of autonomous sailing boats are: long-term planning waterways, depending on weather conditions, short course routing depending on the current data from sensors, collision avoidance and control system for manoeuver execution. Using fuzzy logic for programming the control system for the management of the actuators, has proven to be a very suitable tool for the transfer of expert knowledge of experienced sailors in computer programs in the form of simple IF-THEN rules. The aim of the research, presented in this work, is to propose a new sails and actuators setting, and a management system based on fuzzy logic. We propose the development of an autonomous sailing boat with three different front sails and the possibility of changing the size of the main sail. The possibility of using different sail settings depending on wind and waves conditions at sea, extends the period of high performance (maximum speed) sailing. An autonomous sailing boat with this sail settings, represents a more realistic model of standard sailing boat used for ocean travel. Simulations of a sailing boat with three different front sails and the possibility of changing the size of the main sail, has shown that the use of fuzzy logic in the control system, that operates the actuators, is meaningful and opens the possibility for the development of faster, more stable and energy-efficient vessels.

Z razvojem tehnologije se raziskovalci preizkušajo z izzivi za sestavo sistemov za tvorjenje umetnega govora. Cilj tvorjenja govora je, da na osnovi vhodne informacije v tekstovni obliki, tvorimo signal, ki je razumljiv človeku, seveda pa tako tvorjen govor stremi k čim bolj naravnemu zvoku. Osnova za kvalitetno tvorjenje umetnega govora je ustrezna in kvalitetno obdelana govorna zbirka. Za slovenski jezik obstaja nekaj govornih zbirk, nobena pa ne vključuje govora otroka. Otroški govor je zaradi svoje višje osnovne frekvence specifičen. Prav tako je bolj težavno pridobivanje kvalitetnih posnetkov za govorno zbirko, zaradi manjših psihofizičnih lastnosti otrok in njihove mladoletnosti. Diplomsko delo v prvem delu opisuje postopke za tvorjenje umetnega govora in posebnosti pri tvorjenju slovenskega jezika. V drugem delu pa podaja metode in opis del uporabljenih pri nastanku govorne zbirke MESSIS, ki je prva govorna zbirka otroškega govora v slovenskem jeziku. Zbirka vsebuje 52 min dolg posnetek 7176 besed, govora enega govorca moškega spola. With the development of technology, the researchers are faced with the challenges of building up systems for generating artificial speech. The aim of speech synthesis is that based on the input information in a form of a text, we generate a signal that is understandable to a man. Of course we strive for artificial speech be as natural as possible. Speech database that is adequate and contains data of good quality is essential for generating a high-quality artificial speech. For the Slovenian language, there are some speech databases, none of which includes voice of a child. Children's speech is specific due to its higher main frequency. It is also more difficult to obtain high-quality voice recordings for the collection due to lower physical and mental characteristics of children and their minority. In the first part of the thesis the procedures for generating artificial speech and peculiarities in the formation of the Slovenian language are described. The second part gives a description of the methods used in the creation of speech database MESSIS, the first speech database of a child speech in the Slovenian language. The collection contains 52-minute long sound recording of 7176 words, of one male speaker.

Informacijski sistemi, nas pogosto nezavedno spremljajo na vsakem koraku. Kolikokrat se, ob pogledu na video kamero nad pločnikom, zamislimo in vprašamo, čemu je ta kamera namenjena? Kam se prenaša slika? Kdo sedi za ekranom in spremlja dogajanje? Je slika shranjena? Ali obstaja mehanizem, ki prepozna obraz in ugotavlja, kdo in kdaj hodi po pločniku? S takimi in podobnimi vprašanji, bi se v sodobnem času moral ukvarjati vsakdo. Še posebno se morajo z njimi ukvarjati razvijalci informacijskih sistemov pa najsi gre za strojno ali programsko opremo. Ko položimo prst na biometrično tipalo pri vratih, smo informacijskemu sistemu oddali svoj prstni odtis in z njim enega od osebnih podatkov. Tu uporabniki izgubimo sled o tem osebnem podatku in kaj se z njim v ozadju dogaja. Ne preostane nam drugega, kot da sistemu zaupamo, da bo podatek uporabljen zgolj za avtentikacijo pri odpiranju vrat. Na tem mestu se prične naloga Informacijskega pooblaščenca, ki mora preverjati ali je sistem v katerem se trenutno nahaja naš podatek primerno grajen. Že v fazi razvoja informacijskega sistema mora biti inženir pozoren na vse vidike informacijske varnosti, še posebno pa na varovanje osebnih podatkov in primerno ravnanje z njimi ves čas, ko se podatek nahaja v informacijskem sistemu. Ker so sistemi vedno bolj kompleksni, zakonodaja in standardi zapleteni, si razvijalec lahko pomaga z metodami za presojo vpliva na zasebnost. S to nalogo smo poskusili pomagati inženirju razumeti in izvesti presojo vplivov na zasebnost. At every step and often unconsciously we are being monitored by different information systems. How often do we notice a surveillance camera and ask ourselves what is this camera for? Where is the image transmitted to? Who sits behind the monitor and observes the action? Will my picture be stored? Is behind it an algorithm that recognizes who and when has walked by? These are the questions that everyone should take into consideration. Especially careful should be information system developers of either hardware or software. When we put the finger on the biometric sensor at the door, we have already passed our personal biometric data of the fingerprint to the information system. At this point we lose the track of our personal data and how our data will be processed. There is no other choice but to trust the system that it will process our personal data only for the purpose of opening the door. This is the boundary of the Information Commissioners work, who's duty is to check whether the personal data is being treated accordingly to the law, standards and best practices. Even at the stage of development of the information system, the developing engineer should pay attention to increase the data security to maximum possible level. Special consideration should be taken into account regarding the personal data during the whole data lifetime cycle. Since the systems are becoming more and more complex and so does the legislation and standards, the information system developer can find useful the methods of performing the privacy impact analysis, This paper tries to help the developer to more easily perform the privacy impact assessment.

Predmet diplomskega dela je prototip stroja, ki s pomočjo strojnega vida simulira sortiranje okroglih izdelkov glede na barvo in notranji premer. Prototip vsebuje koračni motor, ki vrti okroglo mizo, le-ta pa ima na eni strani servo motor, ki simulira potiskanje izdelka iz zaloge na mizo. Za prepoznavanje barve in premera skrbita dve spletni kameri. V primeru, da je vzorec primeren, simuliramo potiskanje izdelka iz mize v zaboj z dobrimi izdelki. Za vodenje koračnega motorja in dveh servo motorjev skrbi mikrokrmilnik LPCXpresso 1769, ki s programsko kodo služi kot strežnik z uporabo LwIP TCP/IP kopice. S pomočjo Visual Basic-a, programskega jezika C# in knjižice Emgu CV smo razvili aplikacijo, ki služi kot odjemalec. Le-ta obdeluje vzorce, ki jih zajemamo s spletnima kamerama, in glede na njihovo primernost, preko ethernet povezave, ustrezno javi strežniku, kako naj z vzorci ravna. Za ovrednotenje delovanja stroja smo izdelali kroge različnih barv in različnih premerov sredinskih lukenj in jih postavili na mizo. The subject of this thesis is a prototype machine that uses machine vision to simulate the sorting of round products by color and inner radius. The prototype consists of a stepper motor which rotates a round table, on one side of the prototype is a servo motor, which simulates the pushing of the product from the stock on to the table. Two web cameras are used to identify the color and radius. In the case when the sample is suitable, we simulate the pushing of the product from the table in the crate with good products. For the management of the stepper motor and the two servo motors we use the LPCXpresso 1769 microcontroller that uses the LwIP TCP/IP stack as a server on it. With the use of Visual Basic, the C# programming language and Emgu CV library, we have developed an application that serves as a client that processes the samples that are extracted with the webcams. Therefore, according to their suitability, it appropriately signals the server how to handle the samples via an Ethernet connection. To evaluate the performance of a machine we made circles of different colors and different radius arm holes and laid them on the table.

Namen proizvodne linije je izdelava proizvoda v najkrajšem možnem času z določenimi tolerancami. Načrtovanje le te je prej časovno začrtano zaradi ekonomskega dobička in konkurence sorodnih izdelkov. Nenehno gibanje trga zahteva cenovno prilagajanje in izboljševanje linij ter zagotavljanje nižanja stroška izdelave izdelka. Linija AFS je normirana na 34 sekund na delovno mesto. Izboljšava linije pomeni krajšanje časa oz. ugotavljanje ozkih grl in odpravo le teh. V diplomskem delu sem opisal celotno linijo, prikazal dve ozki grli in način odprave le teh. Idejo smo v podjetju skrbno preučili na marsikaterem sestanku. Za optimizacijo sem dobil nagrado in sem jo delil s sodelavci, ki so mi pomagali pri uresničitvi ideje. Nova nadgradnja linije je z razumno ceno še mogoča, ta pa se konča z nesorazmerjem vložka in izkupička. A production line assumes manufacturing a product in the shortest possible time with certain tolerances. The construction of the line is time planned due to economical profit and the competition of related products. Constant market movements demand cost adjustment and production line improvement that assures product manufacturing cost reduction. The AFS line was normed to 34 seconds per working station. The improvement of the process implies time reduction by acknowledging the line bottlenecks and eliminating them. In my thesis the complete production line is described, I presented two bottlenecks and demonstrated a way of elimination. The idea of solution was thoroughly studied at many meetings. The optimization of the production line rewarded me with a bonus that I shared with my colleagues who contributed to the realization of the idea. At reasonable price, a new production line upgrade is still possible but it unfortunately questions its purpose since the imbalance between input expenses and gained profit is highly noticeable.

Motivacija za zaključno delo je bila želja, da bi spoznali in preizkusili Googlovo aplikacijo za razpoznavanje slovenskega govora. Želja je bila tudi ugotoviti, kako bi bil videti konkreten primer uporabe te aplikacije in na kakšne težave in omejitve bi pri tem naleteli. V prvem poglavju so predstavljena kratka zgodovina in glavna odkritja, ki so prispevala k razvoju sistemov za razpoznavanje govora. Tako lahko vidimo, da je v zadnjih par letih prišlo do velikega napredka pri razpoznavanju govora z uporabo večplastnih nevronskih omrežij. Nevronska omrežja so se v zadnjih letih uveljavila kot orodje za ugotavljanje verjetnosti ujemanja zvočnih signalov s shranjenimi modeli besed oziroma glasov. Tak pristop je prisoten v veliko komercialnih sistemov za razpoznavanje govora. Prav ti sistemi so predstavljeni v drugem poglavju. Tu lahko vidimo, da je na trgu čedalje več razpoznavalnikov govora, razvitih v velikih računalniških podjetjih, kot so Microsoft, Google ali Apple. Ti sistemi se uporabljajo tako za samo narekovanje besedila ali kot ključni del programov osebnih pomočnikov, ki predstavljajo popolnoma nov način interakcije z računalnikom. Iz vsega tega je jasno, da je razvoj razpoznavanja govora v zadnjih letih naredil velik napredek tako v akademski sferi kot v gospodarstvu. Kljub temu obstajajo številne omejitve pri uporabi sistemov za razpoznavanje govora v vsakdanjem življenju. Nekatere od teh problemov spoznamo s testiranjem Googlove aplikacije s pomočjo orodja HResults. Testiranja so bila izvedena na treh različnih skupinah vzorcev govora. Prva skupina je vsebovala vzorce počasnega in razločnega branja, v drugi skupini so bili vzorci hitrejšega in manj razločnega branja, v tretji pa so bili vzorci prostega govora. Točnost razpoznavanja se je razlikovala med različnimi vzorci. Tako je bila povprečna točnost pri razpoznavanju vzorcev iz prve skupine 88-odstotna, pri drugi skupini 79-odstotna in pri tretji 62-odstotna. Rezultati točnosti razpoznavanja pri prvi skupini vzorcev so zelo dobri in so podobni rezultatom, ki jih oglašujejo podjetja ob predstavitvi svojih izdelkov. Iz testov pa lahko vidimo, da ima Googlova aplikacija za razpoznavanje govora težavo pri hitrejšem branju. Težave ima predvsem z razpoznavanjem kratkih besed, kot so vezniki in predlogi, ki jih ne razpozna ali jih razpozna narobe. Pri prostem govoru je razpoznavanje še slabše zaradi velikega števila nepotrebnih glasov in medmetov, tako da so nekateri stavki popolnoma nerazumljivi. Iz rezultatov lahko zaključimo, da je Googlov razpoznavalnik zelo uporaben, če govorimo razločno in ga ne uporabljamo v situacijah, kjer je zahtevana popolna točnost pri razpoznavanju besed. Predstavljeni so tudi rezultati razpoznavanja na večji zbirki vzorcev v dveh različnih obdobjih. Iz preizkusa je razvidno, da se točnost razpoznavanja s časom izboljšuje. V zadnjem poglavju je opisan program, ki je narejen v programskem jeziku Python in je namenjen pošiljanju elektronske pošte s pomočjo govora. Pri razvoju programa lahko spoznamo načine implementacije in delovanje Googlovega sistema za razpoznavanje govora na praktičnem primeru. Googlova aplikacija je izredno zanesljiva, saj ni bilo nobenih problemov s povezavo na njen strežnik. Omejitev časovne dolžine govora, ki ga aplikacija razpozna, predstavlja oviro, če želimo delati z daljšimi zvočnimi datotekami. Ta problem se da rešiti z razdelitvijo zvočnih datotek na krajše dele. Da razdelitev ne uniči govora, jo je treba izvesti na mestih, kjer pride do prekinitve v govoru. Za snemanje program uporablja orodje PyAudio za pretvorbo med zvočnimi formati pa Sound eXchange. Ko imamo transkripcijo celotnega besedila, program besedilo uredi in zamenja narekovana ločila, kot so pika in vejica s pravimi ločili. Na koncu program pošlje besedilo na izbran naslov elektronske pošte. Uporabnost takšnih programov je predvsem v primerih, ko ima uporabnik zasedene roke ali oči z drugim delom in ne more ali ne sme uporabljati tipkovnice. In this thesis, we test the accuracy and usefulness of the Google speech recognition application for the Slovenian language. We also want to find out the limitations and problems of the application. In the first chapter, there is a short overview of history and the most important developments in the field of speech recognition. We can see how deep learning and use of neural networks has profoundly revolutionized automatic speech recognition. Also, a lot of commercial auto speech recognition systems nowadays are based on deep learning methods. Some of these commercial systems are described in the second chapter. The commercial systems for speech recognition have been developed by big corporations, such as Microsoft, Google, and Apple. These systems are a part of the programs that work as intelligent personal assistants. Personal assistants use natural language user interface to answer questions, make recommendations, and perform actions. We tested the accuracy of Google application with the tool HResults. The tests were made on three different groups of samples. The first group included samples of reading that are slow and easy to understand. In the second group, there were samples of faster reading. The third group encompassed speech that is the hardest to understand. The accuracy of recognition was different between different groups of samples. In the first group, recognition accuracy was 88%, recognition accuracy in the second group was 79%, and in the third group 62%. The recognition results are very good for slow reading. These results are similar to results published by advertisements for commercial recognition systems. We can also see that Google application has problems when it tries to recognize faster reading. The recognizer has trouble recognizing short words such as conjunctions and prepositions. When the application recognizes speech that is the hardest to understand, the results are even worse because there are so many unnecessary interruptions and interjections in the speech. We conclude that the Google application is very useful when we speak clearly and we do not use the application in situations where we need perfect accuracy of recognition. We also measure the accuracy of recognition at two different time points. The test shows the accuracy of recognition is improving over time. In the last chapter, we describe the development of our own computer program for sending emails with the help of Google application for speech recognition. Google application is easy to implement into a computer program. The application is also very reliable, but has a limitation on duration of the speech. Time limitation can cause some problems when we want to recognize longer speech files. These problems are solved by cutting audio files into smaller pieces without losing any of the important data. Our program for sending email uses PyAudio for recording audio and it uses Sound eXchange for conversion between different audio formats. When we get the whole transcription of speech from Google application, the program replaces all dictated punctuation marks with real symbols for punctuation marks. At the end, the computer program sends email to the chosen address. These types of programs are useful in situations where the user cannot use a keyboard.

Tema te disertacije je označevanje ujemajočih se parov vzorcev za pridobivanje globinskih slik v zunanjem okolju s pomočjo aktivne strukturirane osvetlitve. Največji problem, s katerim se soočamo pri zajemanju globinskih slik na prostem, so močna osvetlitev ozadja (sonce, drugi svetlobni viri) in motnje drugih podobnih sistemov, ki delujejo na istem območju, ali pa namerne motnje sistema. V delu je predstavljen sistem, ki uporablja poseben, sinhroniziran zajem slike projiciranega vzorca, s katerim odpravi večino motenj in neželene osvetlitve ozadja, s čimer pridobi želeno sliko projiciranega vzorca na prizoru. Pri metodah pridobivanja slike s strukturirano osvetlitvijo je ključni problem določitev ujemajočih se parov točk projiciranega vzorca in točk na zajeti sliki, ki pripadajo projiciranemu vzorcu. Podobni sistemi uporabljajo posebne projekcijske vzorce ali celo barve in s tem olajšajo problem iskanja parov. V tem delu se osredotočimo na iskanje parov z uporabo preprostega enobarvnega vzorca (ena valovna dolžina projicirane svetlobe), ki je sestavljen iz vzporednih premic. Uporaba takega vzorca omogoča uporabo zelo preprostega projektorja vzorcev (laser s posebno lečo, ki žarek razprši v vzporedne premice), uporaba samo ene valovne dolžine svetlobe pa omogoča uporabo ozkopasovnih optičnih filtrov na kameri, ki zajema sliko. S tem se že izloči večina motenj. Pridobljena slika vsebuje projekcijski vzorec na opazovanem prizoru, ki ga je treba izluščiti iz slike in točkam na vzorcu dodeliti oznake oziroma pare, ki so pripadajoče točke na projiciranem vzorcu. V tem delu je predstavljena rešitev, ki ta problem rešuje z uporabo verjetnostnih grafičnih modelov. Predlagamo novo metodo, ki upošteva večje število zaporednih slik, in predstavimo vpliv števila slik na točnost označevanja projiciranih vzorcev. Zaznani projicirani vzorec na sliki razdelimo na večje število rojev slikovnih točk in vsakemu roju dodelimo naključno spremenljivko. Vsi roji slikovnih točk (naključnih spremenljivk) povežemo v verjetnostni grafični model. Naključne spremenljivke, ki pripadajo sosednjim rojem slikovnih točk, uporabimo za izgradnjo faktorjev, s katerimi modeliramo soodvisnost sosednjih naključnih spremenljivk. Upoštevamo soodvisnosti v vodoravni in navpični smeri, vključimo pa še informacijo zaporednih slik, tako da določimo faktorje med prekrivajočimi se roji slikovnih točk v zaporednih slikah na istem položaju na sliki. Izkaže se, da upoštevanje te informacije iz zaporednih slik pomaga pri določevanju oznak rojev slikovnih točk. Kljub specializiranemu senzorju, ki odpravi večino motenj, se lahko v sliki pojavijo slikovni elementi (roji slikovnih točk), motnje, ki niso posledica projiciranega vzorca. Take motnje poskušamo odpraviti s posebno prilagoditvijo grafičnega modela. Naključnim spremenljivkam pri gradnji verjetnostnega grafičnega modela dodamo še eno stanje, ki predstavlja oznako »šum«. Zaželeno je, da model rojem točk, ki ne pripadajo projiciranemu vzorcu, dodeli to oznako. V zadnjem delu predstavimo rešitev z uporabo vrste povratnih nevronskih mrež (ang. recurrent neural network), nevronskega omrežja z dolgim kratkoročnim spominom (ang. Long short-term memory, LSTM). Vhodno sliko razdelimo na večje število manjših področij, ki predstavljajo vhodno zaporedje v nevronsko mrežo. Vsakemu področju dodelimo ustrezno oznako v odvisnosti od dela projekcijskega vzorca na danem področju slike. Nevronsko mrežo učimo tako, da hkrati na sliki zaznava in označuje projicirane vzorce iz surovih podatkov slike brez njene predobdelave. S to metodo je mogoče z veliko natančnostjo na sliki pravilno zaznati in označiti projicirani vzorec. The subject of this dissertation is matching correspondence points for depth image acquisition in outdoor environments utilizing active illumination techniques. The biggest challenge that we are facing when acquiring depth image data outdoor is the presence of strong background illumination (e.g. sunlight) and disturbances caused by other similar systems operating in the same environment or intentional disturbances of the system. In this work a synchronized image acquisition system is presented, which eliminates most of the disturbances and unwanted background illumination, yielding the desired image of illumination pattern on the object. With structured light techniques, the key problem which needs to be solved before obtaining a three dimensional reconstruction of the image is finding pairs of points in the image which correspond to the same part of the projected pattern. Similar active triangulation systems utilize special projection patterns or even color which make the correspondence problem easier. In this work we focus on correspondence problem using a simple monochromatic projection pattern (one wavelength) composed of parallel lines (light planes). Utilizing such a projection pattern enables the use of very simple pattern projectors (laser with a special lens which splits the beam in several parallel lines). The use of monochromatic light also enables the use of band-pass optical filters on the camera, which already filters out most of the disturbances. The acquired frame contains the deformed projected pattern on the observed scene which needs to be segmented out and points on the pattern need to be assigned a pair from the original projected pattern, which comes down to assigning a corresponding label of a projected pattern light plane. In this work a solution utilizing probabilistic graphical models is presented. We propose a method that takes into account a larger number of subsequent frames and test the effect on accuracy of labeling projected patterns. The detected (and segmented out) pattern is divided on a larger number of small pixel-clusters which are also assigned a random variable. All the pixelclusters (random variables) are connected to form a probabilistic graphical model. The random variables which belong to neighboring pixel-clusters are used to form factors through which co-dependencies of neighboring pixel-clusters are modeled. Codependency in vertical and horizontal direction is considered, we also include the information of subsequent images, so that we assign factors between overlapping pixelclusters in subsequent frames on the same position in the image. It turns out that including this information helps with labeling the pixel-clusters. Despite the use of a special imaging sensor which suppresses most of the background and disturbances, unwanted pixel-clusters can appear in the image which are not caused by the projected pattern. We try to remove such pixel-clusters with a special modification of factors in graphical models. When constructing the probabilistic graphical model we assign an additional state to the random variables, which represents the label “noise”. It is desired that the model assigns this label to the pixel-clusters which do not belong to the projected pattern. In the last part of this work we present a solution using a kind of recurrent neural network, long short-term memory (LSTM). The input image is partitioned into a larger number of smaller sub-images, which represent the input sequence for the neural network. Each sub-image is assigned a label depending on the part of the projected pattern which is occupying that particular part of the image. We train the neural network end-to-end from raw image data to label projected patterns in the corresponding parts of the image. The image thus does not need to be preprocessed (thresholding, pixel-cluster detection). Using this method it is possible to detect and assign the correct labels to the projected pattern with high accuracy.

Delo opisuje enega izmed načinov kontrole kakovosti v avtomobilski industriji. Cilj dela je spoznavanje z možnostmi uporabe metod strojnega vida v avtomobilski industriji za razvoj samodejnega zaznavanja slabih nanosov materiala na delih karoserije. Za razvoj in preizkušanje programskih rešitev, ki so predstavljene v tem delu, so bile uporabljene računalniške slike iz dejanske proizvodnje avtomobilov, vendar projekt še ni dosegel take stopnje zrelosti, da bi bil prenesen v praktično uporabo. Na začetku je opisana kontrola kakovosti na splošno in kako je ta uporabljena v industriji. V nadaljevanju je opisan primer strojenega vida iz podjetja Revoz. Sledi predstavitev sistema za nanos materiala in komponent, ki ga sestavljajo. Naštetim načinom trenutnega nadzora nad nepopolnimi nanosi materiala pa sledi opis problemov, ki se kljub temu lahko zgodijo. V tretjem poglavju pa je opisana aplikacija, razvita za zaznavanje nepopolnih nanosov. Opisano je delovanje algoritma in nekaj funkcij, ki ga sestavljajo. Na koncu pa testiramo delovanje aplikacije na primerih slik pomanjkljivih nanosov. Zaradi redkosti napak pri nanosu, ki se pojavljajo med proizvodnim procesom, so bile nekatere slike programsko modificirane v približno tako obliko, kot smo jim bili priča sami. Predstavljajo pa zadosten približek za prikaz delovanja aplikacije. The thesis describes one of the ways to achieve quality assurance in automotive industry. The goal is to familiarise and using machine vision methods to develop automated detection of bad sealant applications on the parts of the car frame. For testing the algorithm and test application we used images from production line because project did not reach the maturity level high enough to be implemented in production process. Starting with general description of quality control and its role in production then continuing with an example of existing system of machine vision used for quality assurance at the company Revoz. In the next subchapter a sealant application system and its components are described. After that we take a look at the already existing methods of detecting bad applications of the sealant followed by some of the problems that happen despite implemented measures. In the third chapter we can see how testing application works and described are functions, settings and an algorithm. And in the last chapter the application and an algorithm are tested using example images from a production line. Situations in which these problems occur are rare that is why the images were manipulated with photo editing software to get examples that look as close as possible to situations witnessed at production process.