Robotics and Computer Engineering - Master's theses

Permanent URI for this collectionhttps://hdl.handle.net/10062/42116

Browse

Browsing Robotics and Computer Engineering - Master's theses by Author "Anbarjafari, Gholamreza, juhendaja"

Filter results by typing the first few letters
Now showing 1 - 5 of 5
  • Thumbnail Image
    Item
    Comprehensive Study on High Dynamic Range Tone Mapping with Subjective Tests
    (Tartu Ülikool, 2017) Salahlı, Aygül; Anbarjafari, Gholamreza, juhendaja; Ozcinar, Cagri, juhendaja; Tartu Ülikool. Loodus- ja täppisteaduste valdkond; Tartu Ülikool. Tehnoloogiainstituut
    A high dynamic range (HDR) image has a very wide range of luminance levels that traditional low dynamic range (LDR) displays cannot visualize. For this reason, HDR images are usually transformed to 8-bit representations, so that the alpha channel for each pixel is used as an exponent value, sometimes referred to as exponential notation [43]. Tone mapping operators (TMOs) are used to transform high dynamic range to low dynamic range domain by compressing pixels so that traditional LDR display can visualize them. The purpose of this thesis is to identify and analyse differences and similarities between the wide range of tone mapping operators that are available in the literature. Each TMO has been analyzed using subjective studies considering different conditions, which include environment, luminance, and colour. Also, several inverse tone mapping operators, HDR mappings with exposure fusion, histogram adjustment, and retinex have been analysed in this study. 19 different TMOs have been examined using a variety of HDR images. Mean opinion score (MOS) is calculated on those selected TMOs by asking the opinion of 25 independent people considering candidates’ age, vision, and colour blindness.
  • Thumbnail Image
    Item
    Design and Comparison of Attitude Control Modes for ESTCube-2
    (Tartu Ülikool, 2017) Ofodile, Ikechukwu Chinonso; Slavinskis, Andris, juhendaja; Anbarjafari, Gholamreza, juhendaja; Tartu Ülikool. Loodus- ja täppisteaduste valdkond; Tartu Ülikool. Tehnoloogiainstituut
    This thesis presents the attitude control problem of ESTCube-2. ESTCube-2 is a 3U CubeSat with a size of 10 x 10 x 30 cm and a weight of about 4 kg. It is the second satellite to be developed by the ESTCube Team and will be equipped with the E-Sail payload for the plasma break experiment, Earth observation camera, a high speed communication system, and a cold gas propulsion module. The satellite will make use of 3 electromagnetic coils, 3 reaction wheels and the cold gas thruster as actuators. The primary purpose of this work was to develop and compare control laws to ful ll the attitude control requirements of the ESTCube-2 mission. To achieve this, the spacecraft dynamics and environmental models are derived and analyzed. PD like controllers and LQR optimal controls are designed to ful ll the pointing requirements of the satellite in addition to the B-dot detumbling control law. Angular rate control law to spin up the satellite for tether deployment is also derived and presented. Simulations of the di erent controllers shows the performance with disturbances also added to the system. Finally recommendations and optimal control situations are presented based on the results.
  • Thumbnail Image
    Item
    Garment retexturing using Kinect V2.0
    (Tartu Ülikool, 2017) Avots, Egils; Anbarjafari, Gholamreza, juhendaja; Escalera, Sergio, juhendaja; Tartu Ülikool. Loodus- ja täppisteaduste valdkond; Tartu Ülikool. Tehnoloogiainstituut
    This thesis describes three new garment retexturing methods for FitsMe virtual fitting room applications using data from Microsoft Kinect II RGB-D camera. The first method, which is introduced, is an automatic technique for garment retexturing using a single RGB-D image and infrared information obtained from Kinect II. First, the garment is segmented out from the image using GrabCut or depth segmentation. Then texture domain coordinates are computed for each pixel belonging to the garment using normalized 3D information. Afterwards, shading is applied to the new colors from the texture image. The second method proposed in this work is about 2D to 3D garment retexturing where a segmented garment of a manikin or person is matched to a new source garment and retextured, resulting in augmented images in which the new source garment is transferred to the manikin or person. The problem is divided into garment boundary matching based on point set registration which uses Gaussian mixture models and then interpolate inner points using surface topology extracted through geodesic paths, which leads to a more realistic result than standard approaches. The final contribution of this thesis is by introducing another novel method which is used for increasing the texture quality of a 3D model of a garment, by using the same Kinect frame sequence which was used in the model creation. Firstly, a structured mesh must be created from the 3D model, therefore the 3D model is wrapped to a base model with defined seams and texture map. Afterwards frames are matched to the newly created model and by process of ray casting the color values of the Kinect frames are mapped to the UV map of the 3D model.
  • Thumbnail Image
    Item
    Local Phase Quantization Feature Extraction based Age and Gender Estimation Using Convolutional Neural Network
    (Tartu Ülikool, 2017) Bilici, Ozan; Anbarjafari, Gholamreza, juhendaja; Tartu Ülikool. Loodus- ja täppisteaduste valdkond; Tartu Ülikool. Tehnoloogiainstituut
    Even though artificial neural networks are one of the oldest machine learning techniques, there were no many experiments on them by 2010s because of its computational complexity. Artificial neural networks got inspired by human neural anatomy, and try to achieve similar accuracy. Latest advances of silicon technology enable us to conduct experiments on all types of artificial neural networks. Convolutional Neural Networks are one of state-of-art neural network types. As a human, we all have great recognition, detection mechanism in our body. In this study, it will be attempted to gain similar ability with computer aid of CNNs. As all other supervisedlearning methods, we need training and testing dataset. We are going to apply CNN on apparent age and gender estimation. There are few public dataset which are created for age estimation. One of them and the biggest one is IMDB-Wiki dataset which contains pictures of famous people from wikipedia and IMDB with their real-age label. In order to create real-age label, the creator used the time differences between photo-taken year and birth year. However for better accuracy, we need apparent age information. Because aging is a process that depends on many conditions. As it is going to be explained later, we collected Japanese dataset on the internet, and labeled their apparent ages by weighted voting. After collecting the image data sets, we pre-processed the images with face detection and alignment methods. Afterwards, we copied all images and used Local Phase Quantization(LPQ) method to extract their features. In CNN, it is always better to use pre-trained data and fine-tune it. Thus we used deep face recognition pre-trained data with almost 2 millions images. After that, we fine tuned images(with LPQ and without LPQ separately) with using the label distribution encoding. Finally we had 2 CNN data. For combining the results, we took the mean of all respective output neurons. At the end, expected values of all neurons are considered as apparent age information. For gender classification, we just trained the system in the similar way. Only difference is that we have only 2 output neurons for gender classification, besides LPQ is not applied in gender classification.
  • Thumbnail Image
    Item
    Semi-Automatic Deflection Measurement Using Digital Image Correlation
    (Tartu Ülikool, 2015) Sundla, Siim; Anbarjafari, Gholamreza, juhendaja; Punning, Andres, juhendaja; Tartu Ülikool. Loodus- ja tehnoloogiateaduskond; Tartu Ülikool. Tehnoloogiainstituut
    The aim of this thesis was to develop an easy-to-use method for providing initial guesses for DIC grid point locations and rotations on objects where major deflection occurs. Study of literature revealed that special approach is required to be able to perform DIC on subsets that are rotated, although such methods exist they are difficult to implement. A method was proposed that employs manual input for creating a curve across image width that describes the specimen deformation. Deflection curve could then be used to provide initial guesses and rotations prior DIC. The approach was implemented using MATLAB and several tests were made to evaluate the algorithm. The results showed that curve based grid transformation can provide sufficient estimation to be able to carry out DIC measurements and converge to sub-pixel accurate solution. The magnitude of displacement error caused by transformation was shown to be irrelevant when search windows are big enough for the correct location to be included. However rotational errors were proven to be significant for sub-pixel accuracy, additional method was proposed and shown to work for compensating errors in rotation estimations. In conclusion the results have shown that the described method can be used for providing initial guesses for grid point locations on deflected objects. Therefore DIC can be used with this approach to carry out deflection measurements.