Department of Electrical and Computer Engineering Graduate Theses
http://hdl.handle.net/1974/794
Sun, 13 Jun 2021 07:39:33 GMT2021-06-13T07:39:33ZAttention and Depth Hallucination for RGB-D Face Recognition with Deep Learning
http://hdl.handle.net/1974/28874
Attention and Depth Hallucination for RGB-D Face Recognition with Deep Learning
Uppal, Hardik
ace recognition approaches that are based purely on RGB images rely solely on intensity information, and therefore are more sensitive to facial variations, notably pose, occlusions, and environmental changes such as illumination and background. These approaches also tend to process the whole image uniformly, weighing distinctive and non-distinctive regions of the image equally. In order to extract more representative facial features, we first propose two fusion techniques based on RGB and depth modalities using attention mechanisms. The first fusion technique uses an LSTM network to selectively focus on feature maps, followed by a convolution layer to generate spatial attention weights. Our method achieves competitive results on CurtinFaces and IIIT-D RGB-D datasets, achieving classification accuracies of over 98.2% and 99.3% respectively. Our second proposed fusion method is a novel attention mechanism that directs the deep network ``where to look'' for visual features in the RGB image by generating an attention map from depth features extracted using a CNN. Our proposed solution achieves notable improvements over the current state-of-the-art on four public datasets, namely Lock3DFace, CurtinFaces, IIIT-D RGB-D, and KaspAROV, with average (increased) accuracies of 87.3% (+5.0%), 99.1% (+0.9%), 99.7% (+0.6%) and 95.3%(+0.5%) for the four datasets respectively, thereby improving the state-of-the-art. Although depth data can provide useful information for face recognition, acquiring depth data in the wild still remains a challenge. To address this problem, we present the Teacher-Student Generative Adversarial Network (TS-GAN) to generate depth images from a single RGB image in order to boost the recognition accuracy of face recognition systems, where depth images are not available. The teacher learns a latent mapping between input RGB and paired depth images in a supervised fashion which the student then generalizes from new RGB data with no available paired depth information. The fully trained shared generator can then be used in runtime to hallucinate depth from RGB for downstream applications such as face recognition. We demonstrate that our hallucinated depth along with the input RGB images boost performance across various architectures when compared to a single RGB modality by average values of +1.2%, +2.6%, and +2.6% for IIIT-D, EURECOM, and LFW datasets respectively.
http://hdl.handle.net/1974/28874A Series and Parallel Tuned DC-DC Resonant Converter with Hybrid Modulation
http://hdl.handle.net/1974/28821
A Series and Parallel Tuned DC-DC Resonant Converter with Hybrid Modulation
Patel, Janki
Series and parallel tuned resonant converters have traditionally been operated at a constant frequency by employing Phase-Shift Modulation (PSM). It provides excellent efficiency from full - load to near no - load while maintaining zero-voltage switching (ZVS). However, the range of input voltage variations is limited. For the applications, where there is a wide variation in the input voltage as well as in the output load, the PSM technique is not able to maintain ZVS, resulting in switching losses. This thesis has studied the behavior and performance of the same converter under hybrid modulation in which both the frequency and the phase-shift are varied. It has been shown that the hybrid modulation technique allows the ZVS operation of the converter under widely varying input voltage and output load.
Frequency-domain modelling is presented, and the performance of the converter is, first, studied under the variable-frequency modulation, and then under the hybrid modulation. It has been shown that the converter can provide better control against the input voltage variations than the fixed frequency modulation by ‘jumping’ the operating frequency from one region to another while maintaining ZVS. The range of the input voltage can be extended further by employing variable pulse-width modulation in addition to variable frequency.
Unlike the constant-frequency operation, the converter exhibits some very interesting characteristics. The new characteristics help in minimizing the conduction losses and allow its operation under a wide range of input voltage and output load.
http://hdl.handle.net/1974/28821Frequency Domain Modelling of Series LC- Parallel LC Resonant Converter with Capacitive Output Filter
http://hdl.handle.net/1974/28796
Frequency Domain Modelling of Series LC- Parallel LC Resonant Converter with Capacitive Output Filter
Mathew, Aiswarya
Multi-element DC-DC resonant converters have been a topic of interest over the past several years. The need for reduced sizes of passive components in the converter, has stimulated substantial research towards the high- frequency operation of these converters, specifically pertaining to high voltage applications. Hence, the series LC-parallel LC (LCLC) resonant converter with capacitive output filter is an emerging favorite for high frequency, high voltage applications. The use of the capacitive output filter ensures reduced peak voltage stress of the rectifier diodes, which is equal to the clamped output voltage. Although the topology exhibits various modes of operation, it predominantly operates in one mode with a lagging power factor. Existing literature utilizes conventional fundamental harmonic approximation (FHA) due to its simplicity. However, FHA has accuracy limitations when applied on low quality factor (Q<2) tuned resonant circuit since higher order harmonics are neglected.
To resolve this issue, this thesis presents an improved frequency domain-based analysis of the LCLC resonant converter with a capacitive output filter, considering the harmonics of both the input and load side of the converter. The proposed analysis can be extended to study converter operational characteristics under variable frequency and fixed- frequency PWM control. Closed- form expressions for the performance characteristics such as voltage gain, resonant tank RMS current, and ZVS angle (β) were derived from the nth harmonic equivalent circuit. The proposed model is validated using simulation studies performed on PSIM (Power Simulation) and by using a hardware prototype. This model improves the accuracy over conventional FHA by 4%.
http://hdl.handle.net/1974/28796COMPREHENSIVE STUDY ON MODULATION OF A DUAL ACTIVE BRIDGE CONVERTER
http://hdl.handle.net/1974/28753
COMPREHENSIVE STUDY ON MODULATION OF A DUAL ACTIVE BRIDGE CONVERTER
Ghazimoghadam, Mohammadali
In recent years, the single-phase Dual Active Bridge (DAB) converter has gained significant importance as it has simple power circuit, bi-directional power flow, and zero voltage switching (ZVS) operation. Although, numerous researchers have worked on this converter and proposed many modulation techniques, a thorough mathematical treatment of this converter has been missing from the literature.
This thesis presents a thorough mathematical treatment of the DAB converter. All the possible theoretical modulation patterns for the DAB converter are derived from the mathematical analysis. It is mathematically proven that there are up to 140 distinct switching patterns that can be applied to the converter. It is shown in the thesis that there are only two modulation patterns those result in the optimal performance of the converter in terms of minimum rms current and guaranteed zero-voltage-switching under wide range of operating conditions.
An analytical solution is developed and a step-by-step algorithm to obtain optimal time ratios for the two selected modulation patterns (denoted by Pattern (z) and Pattern (q) in this thesis) is given. The concept of idle time interval is introduced in this thesis. Then it is shown that the optimal modulation is based on pattern (z) with no idle time interval and pattern (q) with idle time interval. An optimal pattern modulation technique based on variable frequency has been introduced. It has been shown that a frequency variation of about 50% around the optimal power level corresponding to the maximum pattern efficiency can further reduce the conduction losses by 10%.
It is also shown how the optimal modulation can be extended to the scenario with reverse power flow as well as with an inverse ratio of voltages. The modulation pattern presented here is the optimal solution for every DAB converter with arbitrary parameters.
http://hdl.handle.net/1974/28753