Choose an application

• Reference Manager
• EndNote
• RefWorks (Direct export to RefWorks)

On the one hand, with the emergence, in particular, of the Internet of Things (IoT) and artificial intelligence, embedded systems are becoming increasingly popular and are used in many applications, including autonomous machines. The use of autonomous machines has revolutionized the industry, with inventions, such as repetitive task robots used in manufacturing and medical sensors in healthcare. Indeed, to name but two advantages, autonomous machines allow for cost efficiency and quality assurance.
On the other hand, the audiovisual world is becoming increasingly demanding, particularly in terms of image quality. Indeed, while 4K resolution is becoming the quality standard, researchers are already setting up increasingly complex systems to broadcast 8K content on a large scale. This implies a furious growth in the amount of data transferred, given the growth in image quality and the numerous users enjoying video content every single day.
However, the internet network responsible for data transfer is a very limited resource unable to support more and more data without problems. Video streaming is governed by a trade-off between video quality, computation, and compression rate. To succeed in providing ever higher image quality, video content must, therefore, be both compressed at never before achieved compression rates thanks to increasingly complex video encoding standards, and processed by increasingly powerful machines.
The hardware, which is at the center of this work, was the NVIDIA Jetson Xavier AGX developer kit. This developer kit allows to build and produce software solutions on a large scale in Jetson Xavier AGX modules, often used in autonomous machines. Therefore, the purpose of this thesis was to provide a broadcasting solution as powerful and portable as possible on an NVIDIA Jetson Xavier AGX module.
This thesis addresses the issue of video streaming by analyzing the different main video formats, video coding standards, and key video compression techniques. Then, this project deals mainly with the computational part of the trade-off, i.e., providing more computing power to accelerate the decompression of encoded data on embedded systems, via well-known and advanced formats such as H.264 or H.265. For the practical part of this thesis, we will first analyze the decoding performance of a solution based on the specific hardware of the AGX module. Then, we will extend this powerful solution to make it as portable as possible by using CPU and then GPU programming. Finally, we will compare the results provided by the different solutions. D'une part, avec l'émergence, en particulier, de l'Internet of Things (IoT) et de l'intelligence artificielle, les systèmes embarqués sont de plus en plus populaires et sont utilisés dans de nombreuses applications, y compris dans le domaine des machines autonomes. L'utilisation de machines autonomes a révolutionné l'industrie avec des inventions telles que les robots à tâches répétitives, pour la fabrication, et les capteurs médicaux, pour les soins de santé. En effet, pour ne citer que deux avantages, les machines autonomes permettent de réduire les coûts et d'assurer la qualité.
D'autre part, le monde de l'audiovisuel devient de plus en plus exigeant, notamment en termes de qualité d'image. En effet, alors que la résolution 4K est en passe de devenir la norme de qualité, les chercheurs mettent déjà en place des systèmes de plus en plus complexes pour diffuser des contenus 8K à grande échelle. Cela implique une croissance furieuse de la quantité de données transférées, compte tenu de l'augmentation de la qualité des images et des nombreux utilisateurs profitant chaque jour de contenu vidéo.
Or, le réseau internet chargé du transfert de données est une ressource très limitée, incapable de supporter, sans aucun problème, une quantité de données toujours plus importante. Le streaming vidéo est régi par un compromis entre la qualité vidéo, le calcul et le taux de compression. Pour réussir à fournir une qualité d'image toujours plus élevée, le contenu vidéo doit donc être à la fois compressé à des taux de compression jamais atteints auparavant, grâce à des normes d'encodage vidéo de plus en plus complexes, et traité par des machines de plus en plus puissantes.
Le matériel, qui est au centre de ce travail, est le kit de développement NVIDIA Jetson Xavier AGX. Ce kit de développement permet de construire et de produire des logiciels à grande échelle dans les modules Jetson Xavier AGX, souvent utilisés dans les machines autonomes.
L'objectif de cette thèse était donc de fournir une solution de diffusion, aussi puissante et portable que possible, sur un module NVIDIA Jetson Xavier AGX.
Cette thèse aborde la question du streaming vidéo en analysant les principaux formats de vidéo, les normes d'encodage vidéo et les principales techniques de compression vidéo. Ensuite, ce projet traite principalement de la partie calcul du compromis, i.e., en fournissant plus de puissance de calcul pour accélérer la décompression des données encodées sur les systèmes embarqués via des formats connus et avancés, tels que H.264 ou H.265. Pour la partie pratique de cette thèse, nous analyserons d'abord les performances de décodage d'une solution basée sur le matériel spécifique du module AGX. Ensuite, nous étendrons cette solution puissante pour la rendre aussi portable que possible en utilisant la programmation par CPU puis par GPU. Enfin, nous comparerons les résultats fournis par les différentes solutions.

NVIDIA --- Jetson --- Xavier --- AGX --- H.264 --- H.265 --- AVC --- HEVC --- multimedia --- RGB --- YUV --- DCT --- GPU --- CUDA --- Video-streaming --- Video-compression --- Video-decoding --- Parallel-programming --- NDI --- MPEG-2 --- MPEG-4 --- VP8 --- VP9 --- FFmpeg --- Ingénierie, informatique & technologie > Sciences informatiques

Choose an application

• Reference Manager
• EndNote
• RefWorks (Direct export to RefWorks)

The recent development in wireless networks and devices has led to novel services that will utilize wireless communication on a new level. Much effort and resources have been dedicated to establishing new communication networks that will support machine-to-machine communication and the Internet of Things (IoT). In these systems, various smart and sensory devices are deployed and connected, enabling large amounts of data to be streamed. Smart services represent new trends in mobile services, i.e., a completely new spectrum of context-aware, personalized, and intelligent services and applications. A variety of existing services utilize information about the position of the user or mobile device. The position of mobile devices is often achieved using the Global Navigation Satellite System (GNSS) chips that are integrated into all modern mobile devices (smartphones). However, GNSS is not always a reliable source of position estimates due to multipath propagation and signal blockage. Moreover, integrating GNSS chips into all devices might have a negative impact on the battery life of future IoT applications. Therefore, alternative solutions to position estimation should be investigated and implemented in IoT applications. This Special Issue, “Smart Sensor Technologies for IoT” aims to report on some of the recent research efforts on this increasingly important topic. The twelve accepted papers in this issue cover various aspects of Smart Sensor Technologies for IoT.

Internet of Things (IoT) --- ReRoute --- Multicast Repair (M-REP) --- internet of things (IoT) --- Fast Reroute --- bit repair (B-REP) --- failure repair --- WSN --- MANET --- DRONET --- multilayered network model --- 5G --- IoT --- smart sensors --- smart sensor --- IoT system --- Velostat --- pressure sensor --- convolutional neural network --- data classification --- position detection --- magnetometer --- traffic --- vehicle --- classification --- measurement --- detection --- Internet of Things --- Bluetooth --- indoor tracking --- mobile localization --- optical sensors --- vibration sensing --- quality of service differentiation --- wireless optical networks --- free space optics --- multiwavelength laser --- optical code division multiple access (OCDMA) --- underwater wireless sensor network --- energy-efficient --- clustering --- depth-based routing --- mm-wave radars --- GNSS-RTK positioning --- wireless technology --- electromagnetic scanning --- point cloud --- localization --- IMU --- Wi-Fi --- positioning --- dead reckoning --- particle filter --- fingerprinting --- Wi-Fi sensing --- human activity recognition --- location-independent --- meta learning --- metric learning --- few-shot learning --- ACR --- H.264/AVC --- H.265/HEVC --- QoE --- subjective assessment --- n/a

Choose an application

• Reference Manager
• EndNote
• RefWorks (Direct export to RefWorks)

The recent development in wireless networks and devices has led to novel services that will utilize wireless communication on a new level. Much effort and resources have been dedicated to establishing new communication networks that will support machine-to-machine communication and the Internet of Things (IoT). In these systems, various smart and sensory devices are deployed and connected, enabling large amounts of data to be streamed. Smart services represent new trends in mobile services, i.e., a completely new spectrum of context-aware, personalized, and intelligent services and applications. A variety of existing services utilize information about the position of the user or mobile device. The position of mobile devices is often achieved using the Global Navigation Satellite System (GNSS) chips that are integrated into all modern mobile devices (smartphones). However, GNSS is not always a reliable source of position estimates due to multipath propagation and signal blockage. Moreover, integrating GNSS chips into all devices might have a negative impact on the battery life of future IoT applications. Therefore, alternative solutions to position estimation should be investigated and implemented in IoT applications. This Special Issue, “Smart Sensor Technologies for IoT” aims to report on some of the recent research efforts on this increasingly important topic. The twelve accepted papers in this issue cover various aspects of Smart Sensor Technologies for IoT.

Technology: general issues --- Internet of Things (IoT) --- ReRoute --- Multicast Repair (M-REP) --- internet of things (IoT) --- Fast Reroute --- bit repair (B-REP) --- failure repair --- WSN --- MANET --- DRONET --- multilayered network model --- 5G --- IoT --- smart sensors --- smart sensor --- IoT system --- Velostat --- pressure sensor --- convolutional neural network --- data classification --- position detection --- magnetometer --- traffic --- vehicle --- classification --- measurement --- detection --- Internet of Things --- Bluetooth --- indoor tracking --- mobile localization --- optical sensors --- vibration sensing --- quality of service differentiation --- wireless optical networks --- free space optics --- multiwavelength laser --- optical code division multiple access (OCDMA) --- underwater wireless sensor network --- energy-efficient --- clustering --- depth-based routing --- mm-wave radars --- GNSS-RTK positioning --- wireless technology --- electromagnetic scanning --- point cloud --- localization --- IMU --- Wi-Fi --- positioning --- dead reckoning --- particle filter --- fingerprinting --- Wi-Fi sensing --- human activity recognition --- location-independent --- meta learning --- metric learning --- few-shot learning --- ACR --- H.264/AVC --- H.265/HEVC --- QoE --- subjective assessment --- Internet of Things (IoT) --- ReRoute --- Multicast Repair (M-REP) --- internet of things (IoT) --- Fast Reroute --- bit repair (B-REP) --- failure repair --- WSN --- MANET --- DRONET --- multilayered network model --- 5G --- IoT --- smart sensors --- smart sensor --- IoT system --- Velostat --- pressure sensor --- convolutional neural network --- data classification --- position detection --- magnetometer --- traffic --- vehicle --- classification --- measurement --- detection --- Internet of Things --- Bluetooth --- indoor tracking --- mobile localization --- optical sensors --- vibration sensing --- quality of service differentiation --- wireless optical networks --- free space optics --- multiwavelength laser --- optical code division multiple access (OCDMA) --- underwater wireless sensor network --- energy-efficient --- clustering --- depth-based routing --- mm-wave radars --- GNSS-RTK positioning --- wireless technology --- electromagnetic scanning --- point cloud --- localization --- IMU --- Wi-Fi --- positioning --- dead reckoning --- particle filter --- fingerprinting --- Wi-Fi sensing --- human activity recognition --- location-independent --- meta learning --- metric learning --- few-shot learning --- ACR --- H.264/AVC --- H.265/HEVC --- QoE --- subjective assessment

Choose an application

• Reference Manager
• EndNote
• RefWorks (Direct export to RefWorks)

For many decades, researchers have been trying to make computers’ analysis of images as effective as the system of human vision is. For this purpose, many algorithms and systems have previously been created. The whole process covers various stages, including image processing, representation and recognition. The results of this work can be applied to many computer-assisted areas of everyday life. They improve particular activities and provide handy tools, which are sometimes only for entertainment, but quite often, they significantly increase our safety. In fact, the practical implementation of image processing algorithms is particularly wide. Moreover, the rapid growth of computational complexity and computer efficiency has allowed for the development of more sophisticated and effective algorithms and tools. Although significant progress has been made so far, many issues still remain, resulting in the need for the development of novel approaches.

Information technology industries --- Computer science --- CIELab --- component Substitution --- Pan sharpening --- Pléiades VHR Image --- coal --- inertinite macerals --- classification --- multifractal analysis --- support vector machine --- block-based coding --- video coding --- H.265/HEVC --- affine motion compensation --- image registration --- homography matrix --- local homography transformation --- convolutional neural network --- moving direct linear transformation --- super-resolution (SR) --- convolution neural network (CNN) --- Gene Expression Programming (GEP) --- deep learning --- image preclassification --- suspicious behavior detection --- motion --- magnitude --- gradient --- reactivity --- saliency --- haze removal --- dark channel --- atmospheric-light estimation --- coarse-to-fine search strategy --- sparse dictionary --- stable recovery --- frame --- RIP --- local dimming --- retinex theory --- bi-histogram equalization --- contrast ratio --- details preservation --- pansharpening --- image fusion --- image quality --- Satellite Pour l'Observation de la Terre (SPOT) 6 --- spectral consistency --- spatial consistency --- synthesis --- artificial intelligence --- dental application --- images --- detection --- parseval frame --- transform --- sparse representation --- octave convolution --- bilingual scene text reading --- Ethiopic script --- attention --- nasal cytology --- automatic cell segmentation --- rhinology --- image analysis --- feature extraction --- shape context --- plant recognition --- DPCNN --- BOF --- numeral spotting --- historical document analysis --- convolutional neural networks --- deep transfer learning --- handwritten digit recognition --- spectrum correction --- intensity correction --- compressed sensing --- tradeoff process --- IKONOS --- remote sensing --- fine-tuning --- learning rate scheduler --- cyclical learning rates --- label smoothing --- classification accuracy --- neural networks --- salient object detection --- RGB-D --- object detection --- small object --- multi-scale sampling --- balanced sampling --- texture --- structure --- optical --- coke --- iron ore --- sinter --- image processing --- segmentation --- identification --- action recognition --- silhouette sequences --- shape features --- ambient assisted living --- active ageing --- CIELab --- component Substitution --- Pan sharpening --- Pléiades VHR Image --- coal --- inertinite macerals --- classification --- multifractal analysis --- support vector machine --- block-based coding --- video coding --- H.265/HEVC --- affine motion compensation --- image registration --- homography matrix --- local homography transformation --- convolutional neural network --- moving direct linear transformation --- super-resolution (SR) --- convolution neural network (CNN) --- Gene Expression Programming (GEP) --- deep learning --- image preclassification --- suspicious behavior detection --- motion --- magnitude --- gradient --- reactivity --- saliency --- haze removal --- dark channel --- atmospheric-light estimation --- coarse-to-fine search strategy --- sparse dictionary --- stable recovery --- frame --- RIP --- local dimming --- retinex theory --- bi-histogram equalization --- contrast ratio --- details preservation --- pansharpening --- image fusion --- image quality --- Satellite Pour l'Observation de la Terre (SPOT) 6 --- spectral consistency --- spatial consistency --- synthesis --- artificial intelligence --- dental application --- images --- detection --- parseval frame --- transform --- sparse representation --- octave convolution --- bilingual scene text reading --- Ethiopic script --- attention --- nasal cytology --- automatic cell segmentation --- rhinology --- image analysis --- feature extraction --- shape context --- plant recognition --- DPCNN --- BOF --- numeral spotting --- historical document analysis --- convolutional neural networks --- deep transfer learning --- handwritten digit recognition --- spectrum correction --- intensity correction --- compressed sensing --- tradeoff process --- IKONOS --- remote sensing --- fine-tuning --- learning rate scheduler --- cyclical learning rates --- label smoothing --- classification accuracy --- neural networks --- salient object detection --- RGB-D --- object detection --- small object --- multi-scale sampling --- balanced sampling --- texture --- structure --- optical --- coke --- iron ore --- sinter --- image processing --- segmentation --- identification --- action recognition --- silhouette sequences --- shape features --- ambient assisted living --- active ageing

Choose an application

• Reference Manager
• EndNote
• RefWorks (Direct export to RefWorks)

A huge amount of data is acquired nowadays by different remote sensing systems installed on satellites, aircrafts, and UAV. The acquired data then have to be transferred to image processing centres, stored and/or delivered to customers. In restricted scenarios, data compression is strongly desired or necessary. A wide diversity of coding methods can be used, depending on the requirements and their priority. In addition, the types and properties of images differ a lot, thus, practical implementation aspects have to be taken into account. The Special Issue paper collection taken as basis of this book touches on all of the aforementioned items to some degree, giving the reader an opportunity to learn about recent developments and research directions in the field of image compression. In particular, lossless and near-lossless compression of multi- and hyperspectral images still remains current, since such images constitute data arrays that are of extremely large size with rich information that can be retrieved from them for various applications. Another important aspect is the impact of lossless compression on image classification and segmentation, where a reasonable compromise between the characteristics of compression and the final tasks of data processing has to be achieved. The problems of data transition from UAV-based acquisition platforms, as well as the use of FPGA and neural networks, have become very important. Finally, attempts to apply compressive sensing approaches in remote sensing image processing with positive outcomes are observed. We hope that readers will find our book useful and interesting

Technology: general issues --- on-board data compression --- CCSDS 123.0-B-2 --- near-lossless hyperspectral image compression --- hyperspectral image coding --- graph filterbanks --- integer-to-integer transforms --- graph signal processing --- compact data structure --- quadtree --- k2-tree --- k2-raster --- DACs --- 3D-CALIC --- M-CALIC --- hyperspectral images --- fully convolutional network --- semantic segmentation --- spectral image --- tensor decomposition --- HEVC --- intra coding --- JPEG 2000 --- high bit-depth compression --- multispectral satellite images --- crop classification --- Landsat-8 --- Sentinel-2 --- Elias codes --- Simple9 --- Simple16 --- PForDelta --- Rice codes --- hyperspectral scenes --- hyperspectral image --- lossy compression --- real time --- FPGA --- PCA --- JPEG2000 --- EBCOT --- multispectral --- hyperspectral --- CCSDS --- FAPEC --- data compression --- transform --- hyperspectral imaging --- on-board processing --- GPU --- real-time performance --- UAV --- parallel computing --- remote sensing --- image quality --- image classification --- visual quality metrics --- spectral–spatial feature --- multispectral image compression --- partitioned extraction --- group convolution --- rate-distortion --- compressed sensing --- invertible projection --- coupled dictionary --- singular value --- task-driven learning --- on board compression --- transform coding --- learned compression --- neural networks --- variational autoencoder --- complexity --- real-time compression --- on-board compression --- real-time transmission --- UAVs --- compressive sensing --- synthetic aperture sonar --- underwater sonar imaging --- remote sensing data compression --- lossless compression --- compression impact --- computational complexity