Arvutitehnika bakalaureusetööd - Bachelor's theses

Permanent URI for this collectionhttps://hdl.handle.net/10062/42115

Browse

Browsing Arvutitehnika bakalaureusetööd - Bachelor's theses by Author "Anbarjafari, Gholamreza"

Filter results by typing the first few letters
Now showing 1 - 3 of 3
  • Thumbnail Image
    Item
    Edge information based object detection and classification
    (Tartu Ülikool, 2016) Tarvas, Karl; Anbarjafari, Gholamreza; Rasti, Pejman; Tartu Ülikool. Loodus- ja täppisteaduste valdkond; Tartu Ülikool. Tehnoloogiainstituut
    This thesis presents work regarding the development a computationally cheap and reliable edge information based object detection and classification system for use on the NAO humanoid robots. The work covers ground detection, edge detection, edge clustering and cluster classification, the latter task being equivalent to object recognition. Numerous novel improvements are proposed, including a new geometric model for ground detection, a joint edge model using two edge detectors in unison for improved edge detection, and a hybrid edge clustering model. Also, a classification model is outlined along with example classifiers and used values. The work is illustrated graphically where applicable.
  • Thumbnail Image
    Item
    Embedded system for real-time emotional arousal classification
    (Tartu Ülikool, 2020) Rodionov, Kirill; Anbarjafari, Gholamreza; Tartu Ülikool. Loodus- ja täppisteaduste valdkond; Tartu Ülikool. Tehnoloogiainstituut
    We, humans, can distinguish the emotions of others with ease and we always expect any sort of emotional response during a conversation. Machines, however, do not possess emotion related skills, which makes human-machine interactions feel alien and soulless. Therefore, development of an efficient emotion recognition system is one of the crucial steps towards human-like artificial intelligence. A common person can also find use in emotion recognition. It would be a great help to the people, who by various reason either have weak control over own emotions or devoid of any ability to perceive emotions of others. This thesis focuses on creating a solution based on compact hardware to classify emotions in relation to its level of arousal. For this, theory concerning the emotions and their classifications were gathered, after which numerous methods of machine learning and feature description were reviewed and tried out. The methods list support vector machines, random forests, facial landmark feature extraction and histogram of oriented gradients. The project has came to a halt halfway through due to poor results: small scale hardware appeared unsuitable for extensive machine learning operations. It can be resumed with the possibility of introducing another set of hardware purely for recognition models training and leaving the compact one deal with pre-made model. In estonian: Me, inimesed, oskame kergelt tajuda teiste emotsioone, ning ootame mingi emotsionaalset tagasisidet suhtlemise korral. Masinad, kuid, ei oma emotsioonidega seotud oskust, mistõttu inimese ja masina vastastikmõju tundub hingetu ja võõrana. Seepärast, tõhusa emotsiooni tunnustamise arendus on üks ülioluline samm inimesesarnase tehisintellekti suuna. Tava inimene ka saab leida kasu emotsiooni tunnustamises. See saab aidata inimesi, kellel on erinevate põhjuste tõttu nõrk kontroll oma emotsioonide üle või nad ei saa teiste emotsioone tundma. Käesolev töö keskendub kompaktse riistvara baseeritud lahenduse peale emotsiooni liigitamiseks sõltuvalt temast erutusest. Selleks, emotsiooni puudutav teooria oli kogutud, mille pärast arvukad masinõppimise ja tunnuste ekstraheerimise meetodid olid vaadeldatud ja ära proovitud. Need meetodid on tugivektor-masinad, otsustusmetsad, näoorientiiri tunnuste ekstraheerimine ja suunatud gradientide histogramm. Kehva tulemuste tõttu projekt jäi seisma: väikese mastaabi riistvara kujunes vimetuks laiaulatusliku masinõppimise sooritamise jaoks. Seda saab jätkata, kui lisada projekti võimeka riistvara, et ta treeniks tajumiste muudelit ja edastaks kompaktsele riistvarale juba eeltreenitud muudelit rakendamiseks.
  • Thumbnail Image
    Item
    Real-Time Ensemble Based Face Recognition System for Humanoid Robots
    (Tartu Ülikool, 2016) Samuel, Kadri; Anbarjafari, Gholamreza; Bolotnikova, Anastasia; Tartu Ülikool. Loodus- ja täppisteaduste valdkond; Tartu Ülikool. Tehnoloogiainstituut
    Humanoid robots are being used in many industrial and domestic application in which human-robot interaction plays an important role. One of the important existing challenges is developing an accurate real-time face recognition system which is not required to be computationally expensive. In this research work a real-time face recognition system which requires low computational complexity is proposed. For this purpose, this thesis is investigating block processing of local binary patterns of the face images captured by NAO robot, a humanoid. For test purposes, the proposed method is adopted on NAO robot and tested under realworld conditions. The experimental results through this thesis are showing that the proposed face recognition algorithm compares favorably to the conventional and state-of-the-art techniques.