Due to hardware limitations, 3D sensors like LiDAR often produce sparse and noisy point clouds. Point cloud upsampling is the task of converting such point clouds into dense and clean ones. This thesis tackles the problem of point cloud upsampling using deep neural networks. The effectiveness of a point cloud upsampling neural network heavily relies on the upsampling module and the feature extractor used therein. In this thesis, I propose a novel point upsampling module, called NodeShuffle. NodeShuffle leverages Graph Convolutional Networks (GCNs) to better encode local point information from point neighborhoods. NodeShuffle is versatile and can be incorporated into any point cloud upsampling pipeline. Extensive experiments show how NodeShuffle consistently improves the performance of previous upsampling methods. I also propose a new GCN-based multi-scale feature extractor, called Inception DenseGCN. By aggregating features at multiple scales, Inception DenseGCN learns a hierarchical feature representation and enables further performance gains. I combine Inception DenseGCN with NodeShuffle into the proposed point cloud upsampling network called PU-GCN. PU-GCN sets new state-of-art performance with much fewer parameters and more efficient inference.

Thermal cameras are used in numerous computer vision applications, such as human detection and scene understanding. However, the cost of high quality and high resolution thermal sensors is often a limiting factor. Conversely, high resolution visual spectrum cameras are readily available and generally inexpensive. Herein, we explore the creation of higher quality upsampled thermal imagery using a high resolution visual spectrum camera and Markov random fields theory. This paper also presents a discussion of the tradeoffs from this approach and the effects of upsampling, both from quantitative and qualitative perspectives. Our results demonstrate the successful application of this approach for human detection and the accurate propagation of thermal measurements within images for more general tasks like scene understanding. A tradeoff analysis of the costs related to performance as the resolution of the thermal camera decreases are also provided.

Point clouds are a widely used system for the collection and application of 3D data. But most timesthe data gathered is too scarce to reliably be used in any application. Therefore this thesis presentsa GAN based upsampling method within a patch based approach together with a GCN based featureextractor, in an attempt to enhance the density and reliability of point cloud data. Our approachis rigorously compared with existing methods to compare the performance. The thesis also makescorrelations between input sizes and how the quality of the inputs affects the upsampled result. TheGAN is also applied to real-world data to assess the viability of its current state, and to test how it isaffected by the interference that occurs in an unsupervised scenario.

Graph convolutional networks (GCNs) showed promising results in learning from point cloud data. Applications of GCNs include point cloud classification, point cloud segmentation, point cloud upsampling, and more. Recently, the introduction of Deep Graph Convolutional Networks (DeepGCNs) allowed GCNs to go deeper, and thus resulted in better graph learning while avoiding the vanishing gradient problem in GCNs. By adapting impactful methods from convolutional neural networks (CNNs) such as residual connections, dense connections, and dilated convolutions, DeepGCNs allowed GCNs to learn better from non-Euclidean data. In addition, deep learning methods proved very effective in the task of point cloud upsampling. Unlike traditional optimization-based methods, deep learning-based methods to point cloud upsampling does not rely on priors nor hand-crafted features to learn how to upsample point clouds. In this thesis, I discuss the impact and show the performance results of DeepGCNs in the task of point cloud part segmentation on PartNet dataset. I also illustrate the significance of using GCNs as upsampling modules in the task of point cloud upsampling by introducing two novel upsampling modules: Multi-branch GCN and Clone GCN. I show quantitatively and qualitatively the performance results of our novel and versatile upsampling modules when evaluated on a new proposed standardized dataset: PU600, which is the largest and most diverse point cloud upsampling dataset currently in the literature.

<p>The society expects a global interconected digital communication system offering multimedia services, information on demand and interchange of information with a high data rates and low cost. </p><p>All this can not be realized with the terrestrial nets used nowadays cause it is necessary a high economic inversion to get a competitive capacity to interchange information between server and user. The next generation of satellite must have characteristics which improve the current generation, one important requirement is that the same satellite could make a treatment of the different input signals. With this we can avoid a spent of lots of money and time because we do not need terrestrial stations which modify the signals before the information is sent to the satellite.</p><p>For all this, we need an on board treatment of the information in the satellite. We design a frequency bank reallocation (FBR) network by using a filter bank system. This is the first step of the thesis. After we get FBR we introduce some different input signals and analyze the output, using parameters like symbol error rate and variance.</p><p>One important part in the thesis is the QAM signals used to test our system. For this, we design a modulator and a demodulator of QAM4, 16 and 64, paying more attention in the QAM64, cause is the modulation where more errors can appear due to we have got more possible chances which means more precision in the recovery of the signal.</p>

The society expects a global interconected digital communication system offering multimedia services, information on demand and interchange of information with a high data rates and low cost. All this can not be realized with the terrestrial nets used nowadays cause it is necessary a high economic inversion to get a competitive capacity to interchange information between server and user. The next generation of satellite must have characteristics which improve the current generation, one important requirement is that the same satellite could make a treatment of the different input signals. With this we can avoid a spent of lots of money and time because we do not need terrestrial stations which modify the signals before the information is sent to the satellite. For all this, we need an on board treatment of the information in the satellite. We design a frequency bank reallocation (FBR) network by using a filter bank system. This is the first step of the thesis. After we get FBR we introduce some different input signals and analyze the output, using parameters like symbol error rate and variance. One important part in the thesis is the QAM signals used to test our system. For this, we design a modulator and a demodulator of QAM4, 16 and 64, paying more attention in the QAM64, cause is the modulation where more errors can appear due to we have got more possible chances which means more precision in the recovery of the signal.

The successful revival of three-dimensional (3D) cinema has generated a great deal of interest in 3D video. However, contemporary eyewear-assisted displaying technologies are not well suited for the less restricted scenarios outside movie theaters. The next generation of 3D displays, autostereoscopic multiview displays, overcome the restrictions of traditional stereoscopic 3D and can provide an important boost for 3D television (3DTV). Then again, such displays require scene depth information in order to reduce the amount of necessary input data. Acquiring this information is quite complex and challenging, thus restricting content creators and limiting the amount of available 3D video content. Nonetheless, without broad and innovative 3D television programs, even next-generation 3DTV will lack customer appeal. Therefore simplified 3D video content generation is essential for the medium's success. This dissertation surveys the advantages and limitations of contemporary 3D video acquisition. Based on these findings, a combination of dedicated depth sensors, so-called Time-of-Flight (ToF) cameras, and video cameras, is investigated with the aim of simplifying 3D video content generation. The concept of Time-of-Flight sensor fusion is analyzed in order to identify suitable courses of action for high quality 3D video acquisition. In order to overcome the main drawback of current Time-of-Flight technology, namely the high sensor noise and low spatial resolution, a weighted optimization approach for Time-of-Flight super-resolution is proposed. This approach incorporates video texture, measurement noise and temporal information for high quality 3D video acquisition from a single video plus Time-of-Flight camera combination. Objective evaluations show benefits with respect to state-of-the-art depth upsampling solutions. Subjective visual quality assessment confirms the objective results, with a significant increase in viewer preference by a factor of four. Furthermore, the presented super-resolution approach can be applied to other applications, such as depth video compression, providing bit rate savings of approximately 10 percent compared to competing depth upsampling solutions. The work presented in this dissertation has been published in two scientific journals and five peer-reviewed conference proceedings.  In conclusion, Time-of-Flight sensor fusion can help to simplify 3D video content generation, consequently supporting a larger variety of available content. Thus, this dissertation provides important inputs towards broad and innovative 3D video content, hopefully contributing to the future success of next-generation 3DTV.

Super-resolution is a subfield of computer vision centered around upsampling low-resolution images to a corresponding high-resolution counterpart. This degree project investigates the suitability of a deep learning method for real-time video super-resolution. Following earlier work in the field, we use bicubic interpolation as a baseline for comparison. The deep learning method selected is specifically suited towards real-time super-resolution and consists of a motion compensation network and an upsampling network. The deep learning method and bicubic interpolation are compared by quantitatively evaluating the methods against each other in quality metrics and performance metrics. Suitable quality metrics are selected from earlier works to provide increased comparability of results, namely peak signal-to-noise ratio and structure similarity index. The performance metrics are: number of operations for a single upsampled frame, latency, throughput, and memory requirements. We apply the methods to a highly challenging publicly available dataset specifically engineered towards video super-resolution research. To further investigate the deep learning method, we propose a few modifications and study the effect on the metrics. Our findings show that the deep learning models outperform bicubic interpolation in the quality metrics, while bicubic interpolation outperformed the deep learning models in the performance metrics. We also find no significant quality metric improvement associated with having a motion compensation network for this dataset, suggesting that the dataset might be too complex for the motion compensation network. We conclude that the deep learning method exhibits real-time capabilities as the method has a throughput of around 500 frames per second for full HD super-resolution. Additionally, we show that by modifying the deep learning method, we achieve similar latency as bicubic interpolation without sacrificing throughput or quality. / Superupplösning är ett område inom datorseende centrerat kring att uppsampla lågupplösta bilder till högupplösta motsvarigheter. Detta examensarbete undersöker hur lämplig en specifik djupinlärningsmetod är för superupplösning i realtid. Enligt tidigare forskning använder vi oss av bikubisk interpolering som grund för jämförelse. Den valda djupinlärningsmetoden är speciellt anpassad till superupplösning i realtid och består av ett rörelsekompensationsnätverk och ett uppsamplingsnätverk. Djupainlärningsmetoden och interpoleringsmetoden jämförs genom att kvantitativt utvärdera metoderna mot varandra i kvalitetsmått och prestandamått. Lämpliga kvalitetsmått väljs från tidigare forskning för att ge ökad jämförbarhet mellan resultaten, nämligen maximalt signaltill- brusförhållande och strukturlikhetsindex. Prestandamätvärdena är: antal operationer för en uppsamplad bild, latens, genomströmning och minnesbehov. Vi utvärderar metoderna på ett utmanande allmänt tillgängligt dataset speciellt konstruerat för superupplösningsforskning inom video. För att ytterligare undersöka den djupa inlärningsmetoden föreslår vi några modifieringar och studerar effekten på mätvärdena. Våra resultat visar att djupinlärningsmodellerna överträffar bikubisk interpolering i kvalitetsmåtten, medan bikubisk interpolering överträffar djupinlärningsmodellerna i prestandamåtten. Vi finner inte heller någon signifikant kvalitetsmässig förbättring förknippad med att ha ett rörelsekompensationsnätverk för detta dataset, vilket kan betyda att datasetet är för komplext för rörelsekompensationnätverket. Vi drar slutsatsen att djupainlärningsmetoden uppvisar realtidsfunktioner eftersom metoden har en genomströmning på cirka 500 bilder per sekund för full HD superupplösning. Dessutom visar vi att genom att modifiera djupainlärningsmetoden uppnår vi liknande latens som bikubisk interpolering utan att offra genomströmning eller kvalitet.

Fyzikálně přijatelné osvětlení v reálném čase je často dosaženo použitím aproximací. Současné metody často aproximují globální osvětlení v prostoru obrazu s využitím schopností moderních grafických karet. Dva techniky z této kategorie, screen-space ambient occlusion a screen-space directional occlusion jsou popsány detailněji v této práci. Screen-space directional occlusion je zobecněná verze screen-space ambient occlusion s podporou jednoho difúzního odrazu a závislostí na směrové informaci světla. Hlavním cílem projektu bylo experimentování s těmito metodami. Pro uniformní distribuci náhodných vzorek pro obě metody byla použita Halton sekvence. Pro potlačení šumu je použita bilaterální filtrace, která bere do úvahy geometrické vlastnosti scény. Metody jsou dál zrychleny použitím nižších rozlišení pro výpočet. Rekonstrukce výsledků do původní velikosti pro vytvoření konečného obrazu je realizována pomoci joint bilateral upsamplingu. Kromě metod globálního osvětlení byly v práci použity aj metody pro mapování stínů a HDR osvětlení.

The diploma thesis is focused on design and optimalization of digital octave and fraction-octave band filters. This thesis describe the behavior of filters in systems with fixed point arithmetics and investigate the impact of quantization coefficients for frequency response of filter. Filters, whitch has been designed, are implemented into simple software in C. Designed filters are in accordance with standard IEC 61260.