AI at the network edge promises real-time perception and decision-making in safety-critical domains such as aerial robotics, autonomous vehicles, and 5G-enabled infrastructures. Yet, operating under resource constraints, dynamic, and adversarial conditions exposes edge AI systems to fragility, inefficiency, and security risks that threaten their safe operation. GuardAI, a Horizon Europe project, introduces a framework for resilient and trustworthy edge AI that unites three pillars: adversarial robustness, context-enhanced inference, and security-by-design. Initial project results include a diffusion-based adversarial purification framework optimized for real-time operation, lightweight deep unrolling architectures for LiDAR super-resolution with built-in outlier removal, and robust uncertainty quantification modules to improve confidence calibration. It further develops a context-enhanced inference engine that integrates visual, spatial, and operational context across multi-agent systems, and a risk-aware defense recommender that autonomously selects mitigation strategies based on evolving threat landscapes. Through representative Use Cases, covering monitoring with Unmanned Aerial Vehicle, decentralized 5G network analytics, and secure perception in connected autonomous vehicles, GuardAI demonstrates how robust and adaptive AI can be achieved within stringent edge constraints. Together, these technologies lay the groundwork for a new generation of secure, context-aware, and certifiable AI systems that can be trusted to operate autonomously in the physical world.

This work tackles the challenge of enhancing low-resolution LiDAR sensors for SLAM applications through a novel Deep Unrolling-based Super-Resolution (SR) model. We integrate an outlier removal module to ensure structural integrity while maintaining real-time performance. By leveraging a model-based optimization approach, our method efficiently reconstructs high-resolution point clouds while minimizing computational overhead. The proposed SR model is evaluated within a LiDAR SLAM framework, demonstrating significant improvements in pose estimation accuracy and efficiency compared to state-of-the-art SR methods.

The critical perception capabilities of EdgeAI systems, such as autonomous vehicles, are required to be resilient against adversarial threats, by enabling accurate identification and localization of multiple objects in the scene over time, mitigating their impact. Single-agent tracking offers resilience to adversarial attacks but lacks situational awareness, underscoring the need for multi-agent cooperation to enhance context understanding and robustness. This paper proposes a novel mitigation framework on 3D LiDAR scene against adversarial noise by tracking objects based on least-squares graph on multi-agent adversarial bounding boxes. Specifically, we employ the least-squares graph tool to reduce the induced positional error of each detection’s centroid utilizing overlapped bounding boxes on a fully connected graph via differential coordinates and anchor points. Hence, the multi-vehicle detections are fused and refined mitigating the adversarial impact, and associated with existing tracks in two stages performing tracking to further suppress the adversarial threat. An extensive evaluation study on the real-world V2V4Real dataset demonstrates that the proposed method significantly outperforms both state-of-the-art single and multi-agent tracking frameworks by up to 23.3% under challenging adversarial conditions, operating as a resilient approach without relying on additional defense mechanisms or training parameters.

In this paper we propose a methodology combining Federated Learning (FL) with Cross-view Image Geo-localization (CVGL) techniques. We address the challenges of data privacy and heterogeneity in autonomous vehicle environments by proposing a personalized Federated Learning scenario that allows selective sharing of model parameters. Our method implements a coarse-to-fine approach, where clients share only the coarse feature extractors while keeping fine-grained features specific to local environments. We evaluate our approach against traditional centralized and single-client training schemes using the KITTI dataset combined with satellite imagery. Results demonstrate that our federated CVGL method achieves performance close to centralized training while maintaining data privacy. The proposed partial model sharing strategy shows comparable or slightly better performance than classical FL, offering significant reduced communication overhead without sacrificing accuracy. Our work contributes to more robust and privacy-preserving localization systems for autonomous vehicles operating in diverse environments.

This paper proposes a novel localization framework based on collaborative training or federated learning paradigm, for highly accurate localization of autonomous vehicles. More specifically, we build on the standard approach of KalmanNet, a recurrent neural network aiming to estimate the underlying system uncertainty of traditional Extended Kalman Filtering, and reformulate it by the adapt-then-combine concept to FedKalmanNet. The latter is trained in a distributed manner by a group of vehicles (or clients), with local training datasets consisting of vehicular location and velocity measurements, through a global server aggregation operation. The FedKalmanNet is then used by each vehicle to localize itself, by estimating the associated system uncertainty matrices (i.e, Kalman gain). Our aim is to actually demonstrate the benefits of collaborative training for state estimation in autonomous driving, over collaborative decision-making which requires rich V2X communication resources for measurement exchange and sensor fusion under real-time constraints. An extensive experimental and evaluation study conducted in CARLA autonomous driving simulator highlights the superior performance of FedKalmanNet over state-of-the-art collaborative decision-making approaches, in localizing vehicles without the need of real-time V2X communication.

In this paper, we propose a novel cooperative-based system that facilitates each autonomous vehicle of the swarm to be fully aware of its 5 degrees of freedom (DOF) motion, i.e., 3D translation and 2D rotation, a very important task for autonomous navigation, known also as simultaneous localization and mapping (SLAM). The novelty is that the interconnected vehicles of the swarm share a common collective task: simultaneously estimating self and neighboring vehicles’ 5 DOF by perceiving, transmitting, associating and fusing heterogeneous data, e.g., visual, mechanical, satellite based, etc., relying on different sensor modalities and vehicular communication. The proposed sensor fusion framework is based on the Extended Kalman Filter algorithm, which is reformulated in order to capture cooperative 3D translation and 2D rotation estimation in an alternating fashion. Numerical results using the driving parameters of many cars from CARLA simulator, indicate very promising accuracy in terms of absolute trajectory error.

Considering the high cost of high-resolution Lidar sensors, in this work, a novel Lidar super-resolution method is proposed to improve the performance on numerous autonomous vehicle perception tasks, including that of a Lidar odometer. Specifically, we propose a regularized optimization problem employing a learnable regularizer (neural network) to capture the properties of the data. To efficiently solve this problem, a deep unrolling methodology is proposed, thus forming an interpretable and well-justified deep architecture. Extensive experiments on a real-world lidar odometry application highlight that the proposed model exhibits both superior performance as well as a significantly reduced number of trainable parameters i.e., 99.75% less parameters, as compared to other deep learning methods. The source code used for this work can be found at our repository: repository.

This demo presents a novel Deep Federated Unrolling (FL-DU) super-resolution (SR) approach for enabling low-resolution Lidar-based Simultaneous Localization and Mapping (SLAM) solutions. The proposed system enhances the accuracy of low-cost Lidar sensors via novel explainable by design neural networks and by enabling collaboration between individual vehicles during learning, thereby minimizing the need for costly high-resolution Lidar sensors, leading to significant cost reductions without affecting the SLAM accuracy. Our demo is available on https://www.youtube.com/watcn?v=Fp_nBrD6NiY.

This demo presents the resiliency of our global location awareness approach designed for connected and autonomous vehicles (CAVs) operating under network delays and GPS inaccurate measurements. For studying thoroughly our method, a realistic traffic and network simulation framework stack has been developed in order to simulate vehicular ad-hoc networks (VANETs). The proposed information diffusion based approach significantly improves the localization accuracy despite the challenging environment, enhancing vehicles' situational awareness. Our demo is available on: https://youtu.be/vn6r1g3cQo8

Nowadays, three-dimensional (3D) meshes are widely used in various applications in different areas (e.g., industry, education, entertainment and safety). The 3D models are captured with multiple RGB-D sensors, and the sampled geometric manifolds are processed, compressed, simplified, stored, and transmitted to be reconstructed in a virtual space. These low-level processing applications require the accurate representation of the 3D models that can be achieved through saliency estimation mechanisms that identify specific areas of the 3D model representing surface patches of importance. Therefore, saliency maps guide the selection of feature locations facilitating the prioritization of 3D manifold segments and attributing to vertices more bits during compression or lower decimation probability during simplification, since compression and simplification are counterparts of the same process. In this work, we present a novel deep saliency mapping approach applied to 3D meshes, emphasizing decreasing the execution time of the saliency map estimation, especially when compared with the corresponding time by other relevant approaches. Our method utilizes baseline 3D importance maps to train convolutional neural networks. Furthermore, we present applications that utilize the extracted saliency, namely feature-aware multiscale compression and simplification frameworks.

With the advent of Connected and Autonomous Vehicles (CAVs) comes the very real risk that these vehicles will be exposed to cyber-attacks by exploiting various vulnerabilities. This paper gives a technical overview of the H2020 CARAMEL project (currently in the intermediate stage) in which Artificial Intelligent (AI)-based cybersecurity for CAVs is the main goal. Most of the possible scenarios are considered, by which an adversary can generate attacks on CAVs, such as attacks on camera sensors, GPS location, Vehicle to Everything (V2X) message transmission, the vehicle's On-Board Unit (OBU), etc. The counter-measures to these attacks and vulnerabilities are presented via the current results in the CARAMEL project achieved by implementing the designed security algorithms.

Distributed model-predictive controllers provide a robust way to adjust the acceleration of each platoon vehicle and avoid collisions. This is achieved by transforming the control problem into an iterative, finite-horizon optimization with local constraints. However, the derivation of the global optimal solution is not straightforward. In this paper, first, the consensus cost function is formulated, constrained by minimum distance requirements between the vehicles. Then, the solution is derived via the alternating direction method of multipliers (ADMM), an iterative and robust solver with minimal communication demands. A low-complexity solution is proposed by casting the problem as stochastic control optimization. The developed techniques are evaluated via simulations, where the trajectory of the leading vehicle is generated by an open-source software for autonomous driving (CARLA).

In multidimensional times series generated by sensor recordings of multiple dispersed IoT edge devices, missing measurements are commonplace due to sensing or communication failures, considered a thorny and perplexing problem in a wide range of Industry 4.0 applications. Existing studies for time series imputation focus on developing centralized deep learning approaches, which require massive amounts of data to be uploaded to a central server with adequate computational and power resources for the training of the models, since these approaches are unsuitable for edge and IoT devices characterized by limited computation resources. Different from the current literature, in this study, the time series imputation problem is studied from a federated learning perspective, which is able to surmount the above difficulties. In particular, a novel federated learning approach is proposed, assuming different IoT devices with varying sensing and computational capabilities, that trade-off accuracy with computational/communication/sensing complexity and minimize the operations that need to be performed during training and inferences phase. Furthermore, considering that the main computations are performed on the edge, where the IoT edge devices have limited computational capabilities and power resources, a lightweight yet effective autoencoder-based model is employed to address the examined problem, modified properly to capture the temporal dependencies of the time series data. Extensive evaluation studies with two open datasets have shown that both approaches minimize the data exchanges the need to be made for outperforming centralized approaches in the presence of limited training data.

Cooperative Intelligent Transportation Systems envision the integration of cooperative intelligence as a key operational part of autonomous driving. In this way, a fleet or swarm of Connected and Automated Vehicles collectively coordinates its driving actions in order to maximize its performance. To realize this ambition, vehicles need to be fully location-aware of their surrounding environment, through distributed AI intelligence. Motivated by this requirement, we develop in this paper a distributed cooperative awareness scheme which performs multi-modal fusion of heterogeneous sensor sources along with V2V communication information, using graph Laplacian matrix and Least-Mean-Squares algorithm. The intuition behind our approach is that neighboring vehicles are interested in estimating common positions of other vehicles. We build upon our previous work on global awareness though local information diffusion, and prove that the proposed distributed framework is able to address highly efficient the case of lacking any information about other networked vehicles. More specifically, our approach achieves high enough convergence speed as well as location accuracy. The evaluation study has been performed in CARLA autonomous driving simulator and verifies the proposed method’s benefits over other related solutions.

In this paper, we design distributed multi-modal localization approaches for Connected and Automated vehicles. We utilize information diffusion on graphs formed by moving vehicles, based on Adapt-then-Combine strategies coupled with the Least-Mean-Squares and the Conjugate Gradient algorithms. We treat the vehicular network as an undirected graph, where vehicles communicate with each other by means of Vehicle-to-Vehicle communication protocols. Connected vehicles perform cooperative fusion of different measurement modalities, including location and range measurements, in order to estimate both their positions and the positions of all other networked vehicles, by interacting only with their local neighborhood. The trajectories of vehicles were generated either by a well-known kinematic model, or by using the CARLA autonomous driving simulator. The proposed distributed and diffusion localization schemes significantly reduced the GPS error and do not only converged to the global solution, but they even outperformed it. Extensive simulation studies highlight the benefits of the various methods, which in turn outperform other state of the art approaches. The impact of the network connections and the network latency are also investigated.

3D representations of highly deformable 3D models, such as dynamic 3D meshes, have recently become very popular due to their wide applicability in various domains. This trend inevitably leads to a demand for storage and transmission of voluminous data sets, making the need for the design of a robust and reliable compression scheme a necessity. In this work, we present an approach for dynamic 3D mesh compression, that effectively exploits the spatio-temporal coherence of animated sequences, achieving low compression ratios without noticeably affecting the visual quality of the animation. We show that, on contrary to mainstream approaches that either exploit spatial (e.g., spectral coding) or temporal redundancies (e.g., PCA-based method), the proposed scheme, achieves increased efficiency, by projecting the differential coordinates sequence to the subspace of the covariance of the point trajectories. An extensive evaluation study, using, different dynamic 3D models, highlights the benefits of the proposed approach in terms of both execution time and reconstruction quality, providing extremely low bit-per-vertex- per-frame (bpvf) rates.

Deep learning (DL) tends to be the integral part of Autonomous Vehicles (AVs). Therefore the development of scene analysis modules that are robust to various vulnerabilities such as adversarial inputs or cyber-attacks is becoming an imperative need for the future AV perception systems. In this paper, we deal with this issue by exploring the recent progress in Artificial Intelligence (AI) and Machine Learning (ML) to provide holistic situational awareness and eliminate the effect of the previous attacks on the scene analysis modules. We propose novel multi-modal approaches against which achieve robustness to adversarial attacks, by appropriately modifying the analysis Neural networks and by utilizing late fusion methods. More specifically, we propose a holistic approach by adding new layers to a 2D segmentation DL model enhancing its robustness to adversarial noise. Then, a novel late fusion technique has been applied, by extracting direct features from the 3D space and project them into the 2D segmented space for identifying inconsistencies. Extensive evaluation studies using the KITTI odometry dataset provide promising performance results under various types of noise.

The deployment of high performing deep learning models on platforms of limited resources is currently an active area of research. Among the main directions followed so far, pre-trained neural networks are accelerated and compressed by appropriately modifying their structure and / or parameters. Capitalizing on a recently proposed codebook of a special structure that can be utilized in the frame of the so-called weight sharing methods, this paper describes a "data-driven" technique for designing such a codebook. The performance of the technique, in terms of the observed representation error and classification accuracy versus the achieved acceleration ratio, is demonstrated by considering the VGG16 and the ResNet18 models, pre-trained on the ILSVRC2012 dataset.

The design of 3D object detection schemes that use point clouds as input in automotive applications has gained a lot of interest recently. Those schemes capitalize on Deep Neural Networks (DNNs) that have demonstrated impressive results in analyzing complex scenes. The proposed schemes are generally designed to improve the achieved performance, leading however to high performing approaches with high computational complexity. To mitigate this high complexity and to facilitate their deployment on edge devices, model compression and acceleration techniques can be utilized. In this paper, we propose compressed versions of two well-known 3D object detectors, namely, PointPillars and PV-RCNN, utilizing dictionary learning-based weight-sharing techniques. It is demonstrated that significant acceleration gains can be achieved with acceptable average precision loss when evaluated on the KITTI 3D object detection benchmark. These findings constitute a concrete step towards the deployment of high-performance networks in edge devices of limited resources, such as NVIDIA’s Jetson TX2.

New generation 3-D scanning technologies are expected to create a revolution at the Industry 4.0, facilitating a large number of virtual manufacturing tools and systems. Such applications require the accurate representation of physical objects and/or systems achieved through saliency estimation mechanisms that identify certain areas of the 3-D model, leading to a meaningful and easier to analyze representation of a 3-D object. 3-D saliency mapping is, therefore, guiding the selection of feature locations and is adopted in a large number of low-level 3-D processing applications including denoising, compression, simplification, and registration. In this article, we propose a robust and fast method for creating 3-D saliency maps that accurately identifies sharp and small-scale geometric features in various industrial 3-D models. An extensive experimental study using a large number of 3-D scanned and CAD models verifies the effectiveness of the proposed method as compared to other recent and relevant approaches despite the constraints posed by complex geometry patterns or the presence of noise.

Recent advances in 3-D scanning technology have enabled the deployment of 3-D models in various industrial applications such as digital twins, remote inspection, and reverse engineering. Despite their evolving performance, 3-D scanners still introduce noise and artifacts in the acquired dense models. In this article, we propose a fast and robust denoising method for the dense 3-D scanned industrial models. The proposed approach employs conditional variational autoencoders to effectively filter face normals. Training and inference are performed in a sliding patch setup reducing the size of the required training data and execution times. We conducted extensive evaluation studies using 3-D scanned and CAD models. The results verify plausible denoising outcomes, demonstrating similar or higher reconstruction accuracy, compared to other state-of-the-art approaches. Specifically, for 3-D models with more than 1e4 faces, the presented pipeline is twice as fast as methods with equivalent reconstruction error.

Throughout the years, several works have been proposed for 3D mesh denoising. Nevertheless, despite their reconstruction quality, there are still challenges related to the preservation of the fine surface features. Motivated by the impressive results of image denoising by 3D transform-domain collaborative filtering (CF), we extend it to 3D mesh denoising. CF is also capable of revealing the finest details shared by grouped blocks while preserving at the same time the unique features of each block. A new promising approach suggests unrolling the computational pipeline of CF into a convolutional neural network (CNN) structure increasing significantly the efficiency of this solution. In this paper, we successfully extend and apply this method to 3D meshes making a transition from face normals to pixels. Extensive evaluation studies carried out using a variety of 3D meshes verify that the proposed approach achieves plausible reconstruction outputs and provides very promising results.

Mesh saliency has been widely considered as the measure of visual importance of certain parts of 3D geometries, distinguishable from their surroundings, with respect to human visual perception. This work is based on the use of convolutional neural networks to extract saliency maps for large and dense 3D scanned models. The network is trained with saliency maps extracted by fusing local and global spectral characteristics. Extensive evaluation studies carried out using various 3D models, include visual perception evaluation in simplification and compression use cases. As a result, they verify the superiority of our approach as compared to other state-of-the-art approaches. Furthermore, these studies indicate that CNN-based saliency extraction method is much faster in large and dense geometries, allowing the application of saliency aware compression and simplification schemes in low-latency and energy-efficient systems.

Deep learning and especially the use of Deep Neural Networks (DNNs) provides impressive results in various regression and classification tasks. However, to achieve these results, there is a high demand for computing and storing resources. This becomes problematic when, for instance, real-time, mobile applications are considered, in which the involved (embedded) devices have limited resources. A common way of addressing this problem is to transform the original large pre-trained networks into new smaller models, by utilizing Model Compression and Acceleration (MCA) techniques. Within the MCA framework, we propose a clustering-based approach that is able to increase the number of employed centroids/representatives, while at the same time, have an acceleration gain compared to conventional, k-means based approaches. This is achieved by imposing a special structure to the employed representatives, which is enabled by the particularities of the problem at hand. Moreover, the theoretical acceleration gains are presented and the key system hyper-parameters that affect that gain, are identified. Extensive evaluation studies carried out using various state-of-the-art DNN models trained in image classification, validate the superiority of the proposed method as compared for its use in MCA tasks.

Delineation approaches provide significant benefits to various domains, including agriculture, environmental and natural disasters monitoring. Most of the work in the literature utilize traditional segmentation methods that require a large amount of computational and storage resources. Deep learning has transformed computer vision and dramatically improved machine translation, though it requires massive dataset for training and significant resources for inference. More importantly, energy-efficient embedded vision hardware delivering real-time and robust performance is crucial in the aforementioned application. In this work, we propose a U-Net based tree delineation method, which is effectively trained using multi-spectral imagery but can then delineate single-spectrum images. The deep architecture that also performs localization, i.e., a class label corresponds to each pixel, has been successfully used to allow training with a small set of segmented images. The ground truth data were generated using traditional image denoising and segmentation approaches. To be able to execute the proposed DNN efficiently in embedded platforms designed for deep learning approaches, we employ traditional model compression and acceleration methods. Extensive evaluation studies using data collected from UAV s equipped with multi-spectral cameras demonstrate the effectiveness of the proposed methods in terms of delineation accuracy and execution efficiency.

The Industrial Internet of Things (IIoT) is a key element of industry 4.0, bringing together modern sensor technology, fog and cloud computing platforms, and artificial intelligence to create smart, self-optimizing industrial equipment and facilities. Though, the scale and sensitivity degree of information continuously increases, giving rise to serious privacy concerns. The scope of this article is to provide efficient privacy preservation techniques, by tracking the correlation of multivariate streams recorded in a network of IIoT devices. The time-varying data covariance matrix is used to add noise that cannot be easily removed by filtering, generating obfuscated measurements and, thus, preventing unauthorized access to the original data. To improve communication efficiency between connected IoT devices, we exploit inherent properties of the correlation matrices, and track the essential correlations from a small subset of correlation values. Extensive simulation studies using constrained IIoT devices validate the robustness, efficiency, and effectiveness of our approach.

Chronic inflammatory conditions are obstructive respiratory diseases that affect negatively the quality of life for patients and their families worldwide. The effective control of these diseases is achieved through the use of pressurized meter dose inhaler (pMDI). However, their management has been considered suboptimal, mainly due to the improper use of the inhaler device. Towards this direction, this work presents the use of deep sparse Convolutional Neural Network (CNN) as a classifier to provide a real-time assessment of medication adherence. The classification process is based only on recordings of inhaler use, achieving at the same time significantly lower computational complexity as compared to recent and relevant approaches, since no data transformation is needed. The proposed scheme reaches 95% classification accuracy, demonstrating that this method can also be executed on an embedded system dedicated to medication monitoring. Finally, timing studies were carried out and compared with other classification methods validating its computational efficiency.

Throughout the years, several works have been proposed for performing feature-preserving mesh denoising. Despite their reconstruction benefits, there are still challenges that need to be addressed. Meanwhile, several researchers, system engineers, and software developers have shown increasing interest in the application of deep learning approaches for performing several low-level information processing tasks, such as denoising, compression, etc., in image and video applications. In this paper, we present a method for reconstructing accurately static and dynamic noisy meshes by applying autoencoders on guided normals. To increase the effectiveness of the proposed method, we use different models for different set of faces that are classified as features and non-features. Extensive evaluation studies carried out using a variety of static and dynamic meshes, verify that the proposed approach achieves plausible reconstruction output despite the constraints posed by complex noise and geometry patterns.

Many mixed reality applications are based on the real-time compression and streaming of three-dimensional (3D) models. Thus, they demand very high-bandwidth and ultra-low latency from network specifications. The next-generation wireless networks will employ promising technologies to significantly improve the communication data rates. However, due to implementation complexity and thus increased energy consumption of these technologies, a trade-off between the quality-of-user-experience (QoE) and the hardware specifications is necessary. To overcome these limitations low-resolution quantizers have been of interest, which provide a trade-off between quality and complexity. In this paper, we propose a complexity-aware perceptual coding scheme that minimizes the reconstruction losses of the 3D models. Extensive simulations assuming different 3D models show that the proposed scheme achieves plausible reconstruction output offering significantly higher energy efficiency gains, as compared to a context unaware coding approaches.

The latest advancements in 3D scanning technologies have facilitated the generation and adoption of 3D models in several industrial applications ranging from manufacturing inspection and repair to digital twins and medical industry. All these applications require an accurate representation of physical objects through saliency mechanisms identifying certain areas of the 3D model that are considered as important information by humans. Hence, 3D saliency mapping is an essential mechanism in a number of 3D processing applications including denoising, compression, simplification, registration, viewpoint selection, etc. In this work, we propose a robust 3D saliency mapping method ideally suited for industrial 3D models with sharp and small scale geometric features. An extensive simulation study using a variety of 3D scanned and CAD models, verify the effectiveness of the proposed method as compared to other relevant approaches.

The Industrial IoT (IIoT) is a key element of Industry 4.0, bringing together modern sensor technology, fog - cloud computing platforms, and artificial intelligence (AI) to create smart, self-optimizing industrial equipment and facilities. Though, the scale and sensitivity degree of information continuously increases, giving rise to serious privacy concerns. In this work we address the problem of efficiently and effectively tracking the structure of multivariate streams recorded in a network of IIoT devices. The time varying correlation data values are used to add noise which maximally preserves privacy, in the sense that it is very hard to be removed. To improve communication efficiency between connected IoT devices, we exploit low rank properties of the correlation matrices, and track the essential correlations from a small subset of correlation values estimated by a subset of network nodes. Extensive simulation studies, validate the correctness, efficiency, and effectiveness of our approach in terms of computational complexity, transmission energy efficiency and privacy preservation.

Through the years, several works have demonstrated high-quality 3D mesh denoising. Despite the high reconstruction quality, there are still challenges that need to be addressed ranging from variations in configuration parameters to high computational complexity. These drawbacks are crucial especially if the reconstructed models have to be used for quality check, inspection or repair in manufacturing environments where we have to deal with large objects resulting in very dense 3D meshes. Recently, deep learning techniques have shown that are able to automatically learn and find more accurate and reliable results, without the need for setting manually parameters. In this work, motivated by the aforementioned requirements, we propose a fast and reliable denoising method that can be effectively applied for reconstructing very dense noisy 3D models. The proposed method applies conditional variational autoencoders on face normals. Extensive evaluation studies carried out using a variety of 3D models verify that the proposed approach achieves plausible reconstruction outputs, very relative or even better of those proposed by the literature, in considerably faster execution times.

The increasing interest for reliable generation of large scale scenes and objects has facilitated several real-time applications. Although the resolution of the new generation geometry scanners are constantly improving, the output models, are inevitably noisy, requiring sophisticated approaches that remove noise while preserving sharp features. Moreover, we no longer deal exclusively with individual shapes, but with entire scenes resulting in a sequence of 3D surfaces that are affected by noise with different characteristics due to variable environmental factors (e.g., lighting conditions, orientation of the scanning device). In this work, we introduce a novel coarse-to-fine graph spectral processing approach that exploits the fact that the sharp features reside in a low dimensional structure hidden in the noisy 3D dataset. In the coarse step, the mesh is processed in parts, using a model based Bayesian learning method that identifies the noise level in each part and the subspace where the features lie. In the feature-aware fine step, we iteratively smooth face normals and vertices, while preserving geometric features. Extensive evaluation studies carried out under a broad set of complex noise patterns verify the superiority of our approach as compared to the state-of-the-art schemes, in terms of reconstruction quality and computational complexity.

Recently, there is a tremendous interest in the processing of unorganized point clouds, generated using a variety of 3D scanning technologies such as structured light and LIDAR systems. Without a doubt, the most compelling problem in this domain is the removal of outliers. To effectively address the aforementioned issue, we present a novel method, that detects accurately and efficiently the outliers by exploiting the spatial coherence in the object geometry and the sparsity of the outliers in the spatial domain. This is achieved by solving a convenient convex method called Robust PCA (RPCA). To demonstrate the effectiveness of the proposed technique, we evaluate it by using real scanned point clouds which are extremely dense consisting of millions of points.

There are several works performing accurate skeleton extraction, however, their main drawback is the extensive computational requirements and the lack of solutions that can be executed in multi-core computing systems. These challenges, become more demanding when we are dealing with dense 3D models. To cope with this scarcity we propose a novel method that extends a well known contraction-based skeletonization method, enabling its decentralization resulting in significant improvement in skeleton extraction times.

Recently, there has been increasing interest in the processing of dynamic scenes as captured by 3D scanners, ideally suited for challenging applications such as immersive tele-presence systems and gaming. Despite the fact that the resolution and accuracy of the modern 3D scanners are constantly improving, the captured 3D point clouds are usually noisy with a perceptive percentage of outliers, stressing the need of an approach with low computational requirements which will be able to automatically remove the outliers and create a consolidated point cloud. In this paper, we introduce a novel method which first recognizes and removes outliers from a dynamic point cloud sequence (DPCS) using a very fast Robust PCA (RPCA) approach and then we use a novel weighted Laplacian interpolation approach to achieve a fast and effective consolidation of a DPCS. Extensive evaluation studies, carried out using a collection of different DPCS, verify that the proposed technique achieves plausible reconstruction output despite the constraints posed by arbitrarily complex motion scenarios.

In this paper, we present a progressive compression scheme that enables aggressive compression ratios, by successfully identifying and encoding sharp and small scale geometric features. The accurate identification of the features is achieved by exploiting the low rank property of the captured geometry and the sparsity of the features in the Laplacian domain, permeating benefits from robust principal component analysis. Due to the visual importance of the identified geometric features, the geometry coding process, is optimized for preserving the geometric features at extremely low bit rates. Extensive evaluation studies, carried out using a collection of scanned and synthetic 3D models, show that the proposed feature aware high pass quantization method achieves extremely high compression ratios, offering at the same time meaningful approximations of the given surfaces. Finally, a short discussion regarding the applicability of the proposed feature identification scheme to smooth completion and feature preserving surface denoising is also offered.

Recently, there has been increasing interest for easy and reliable generation of 3-D animated models facilitating several real-time applications (like immersive telepresence, motion capture, and gaming). In most of these applications, the reconstruction of soft body animations is based on time-varying point clouds, which are nonuniformly sampled and highly incomplete. To overcome these significantly challenging imperfections without any additional information, first we introduce a novel reconstruction technique based on rank minimization theory, which can result in a unique solution to the otherwise ill-posed problem. This technique is further extended to exploit the spatial coherence, which usually characterizes the soft-body animations. Based on the developed tools, we propose a distributed consolidation technique where the reconstruction is performed by working simultaneously on several groups of frames. To achieve this, we impose temporal coherence between successive frame clusters by constraining the rank minimization problem. We validate the proposed techniques via experimental evaluation under different configurations and animated models, where we show that the high-frequency details of the models can be adequately recovered from a highly incomplete geometry data set.

Recently, there has been increasing interest for easy and reliable generation of 3D animated models facilitating several real-time applications. In most of these applications, the reconstruction of soft body animations is based on time-varying point clouds which are irregularly sampled and highly incomplete. To overcome these imperfections, we introduce a novel reconstruction technique, using graph-based matrix completion approaches. The presented method exploits spatio-temporal coherences by implicitly forcing the proximity of the adjacent 3D points in time and space. The proposed constraints are modeled by using the weighted Laplacian graphs and are constructed from the available points. Extensive evaluation studies, carried out using a collection of different highly-incomplete dynamic models, verify that the proposed technique achieves plausible reconstruction output despite the constraints posed by arbitrarily complex and motion scenarios.

With the growing demand for easy and reliable generation of 3D models representing real-world or synthetic objects, new schemes for acquisition, storage, and transmission of 3D meshes are required. In principle, 3D meshes consist of vertex positions and vertex connectivity. Vertex position encoders are much more resource demanding than connectivity encoders, stressing the need for novel geometry compression schemes. The design of an accurate and efficient geometry compression system can be achieved by increasing the compression ratio without affecting the visual quality of the object and minimizing the computational complexity. In this paper, we present novel compression/reconstruction schemes that enable aggressive compression ratios, without significantly reducing the visual quality. The encoding is performed by simply executing additions/subtractions. The benefits of the proposed method become more apparent as the density of the meshes increases, while it provides a flexible framework to trade efficiency for reconstruction quality. We derive a novel Bayesian learning algorithm that models the most significant graph Fourier transform coefficients of each submesh, as a multivariate Gaussian distribution. Then we evaluate iteratively the distribution parameters using the expectation-maximization approach. To improve the performance of the proposed approach in highly under determined problems, we exploit the local smoothness of the partitioned surfaces. Extensive evaluation studies, carried out using a large collection of different 3D models, show that the proposed schemes, as compared to the state-of-the-art approaches, achieve competitive compression ratios, offering at the same time significantly lower encoding complexity.

Geometry contraction algorithms are introduced in this work to implement the diverse respiratory configurations of lung related diseases associated with airways obstructions. In addition, computational fluid dynamics (CFD) techniques along with fluid particle tracing (FPT) methods are utilized to efficiently evaluate the behavior of the airflow during the inhalation period, as well as to clarify the features of the inhaled particles in terms of regional deposition. Useful deductions are drawn regarding personalized medication in obstructed conditions.

As large-scale dense and often randomly deployed wireless sensor networks (WSNs) become widespread, local information exchange between colocated sets of nodes may play a significant role in handling the excessive traffic volume. Moreover, to account for the limited life-span of the wireless devices, harvesting the energy of the network transmissions provides significant benefits to the lifetime of such networks. In this paper, we study the performance of communication in dense networks with wireless energy harvesting (WEH)-enabled sensor nodes. In particular, we examine two different communication scenarios (direct and cooperative) for data exchange and we provide theoretical expressions for the probability of successful communication. Then, considering the importance of lifetime in WSNs, we employ state-of-the-art WEH techniques and realistic energy converters, quantifying the potential energy gains that can be achieved in the network. Our analytical derivations, which are validated by extensive Monte-Carlo simulations, highlight the importance of WEH in dense networks and identify the tradeoffs between the direct and cooperative communication scenarios.

As the number of nodes in wireless sensor networks (WSNs) increases, new challenges have to be faced in order to maintain their performance. A fundamental requirement of several applications is the correct transmission of the measurements to their final destinations. Thus, it is crucial to guarantee a high probability of connectivity, which characterizes the ability of every node to report to the fusion center. This network metric is strongly affected by both the fading characteristics and the different routing protocols that are used for the dissemination of data. In this paper, we study the probability of a network to be fully connected for two widely employed routing mechanisms, namely unicast and K-anycast. The analytical derivations and the simulations evaluate the trade-offs among the different routing mechanisms and provide useful guidelines on the design of WSNs.

Wireless body area networks (WBANs) are composed of sensors that either monitor and transmit vital signals or act as relays that forward the received data to a body node coordinator (BNC). In this paper, we introduce an energy efficient vital signal telemonitoring scheme, which exploits compressed sensing (CS) for low-complexity signal compression/reconstruction and distributed cooperation for reliable data transmission to the BNC. More specifically, we introduce a cooperative compressed sensing (CCS) approach, which increases the energy efficiency of WBANs by exploiting the benefits of random linear network coding (RLNC). We study the energy efficiency of RLNC and compare it with the store-and-forward (FW) protocol. Our mathematical analysis shows that the gain introduced by RLNC increases as the link failure rate increases, especially in practical scenarios with a limited number of relays. Furthermore, we propose a reconstruction algorithm that further enhances the benefits of RLNC by exploiting key characteristics of vital signals. With the aid of electrocardiographic (ECG) and electroencephalographic (EEG) data available in medical databases, extensive simulation results are illustrated, which validate our theoretical findings and show that the proposed recovery algorithm increases the energy efficiency of the body sensor nodes by 40% compared to conventional CS-based reconstruction methods.

Wireless Body Area Networks (WBANs) are composed of various sensors that either monitor and transmit real time vital signals or act as relays that forward the received data packets to a nearby Body Node Coordinator (BNC). The design of an accurate and energy efficient wireless telemonitoring system can be achieved by: i) minimizing the amount of data that should be transmitted for an accurate reconstruction at the BNC, and ii) increasing the robustness of the telemonitoring system to link failures due to the nature of wireless medium. To this end, we present a novel Compressed Sensing (CS) based telemonitoring scheme, called Cooperative Compressed Sensing (CCS), that exploits the benefits of Random Linear Network Coding (RLNC) along with key characteristics of the transmitted biosignals in order to achieve an energy efficient signal reconstruction at the BNC. Simulation studies, carried out with real electrocardiographic (ECG) data, show the benefits of: i) employing RLNC, compared to the case where relays simply store and forward the original data packets, and ii) applying the proposed CCS scheme, compared to traditional CS recovery approaches.

Random dense deployment of heterogeneous networks (HetNets), consisting of macro base stations (BS) and small cells (SC), can provide higher quality of service (QoS) while increasing the energy efficiency of the cellular network. In addition, it is possible to achieve lower deployment cost and, therefore, maximize the benefits for the network providers. In this paper, we propose a novel method to determine the minimum deployment cost of a two-tier heterogeneous cellular network using random deployment. After deriving the coverage probability of the two-tier deployment by using stochastic geometry tools, we identify the tier intensities that provide the minimum deployment cost for a given coverage probability. Extensive simulations verify the existence of a unique set of intensities for different coverage constraints.

Life expectancy is projected to increase significantly in the coming years. This fact has pushed the need for designing new and more pervasive healthcare systems. In this field, distributed and networked embedded systems, such as Wireless Sensor Networks (WSNs), are the most suitable technology to achieve continuous monitoring of aged people for their own safety, without affecting their daily activities. This paper proposes recent advancements in this field by introducing WSN4QoL, a Marie Curie project which involves academic and industrial partners from three EU countries. The project aims to propose new WSN-based technologies to meet the specific requirements of pervasive healthcare applications. In particular, in this paper, a Network Coding (NC) mechanism and a distributed localization solution are presented. They have been implemented on WSN testbeds to achieve efficiency in the communications and to enable indoor people tracking. Preliminary results in a real environment show good system performance that meet our expectations.

In this paper novel subspace-based blind schemes are proposed and applied to the sparse channel identification problem. Moreover, adaptive sparse subspace tracking methods are proposed so as to provide efficient real-time implementations. The new algorithms exploit the subspace sparsity either via employing ℓ1-norm relaxation or through greedy-based optimization. The derived schemes have been tested in a Zero-Prefix Orthogonal Frequency Division Multiplexing (ZP-OFDM) system and it turns out that, compared to state-of-art existing schemes, they offer improved performance in terms of convergence rate and steady-state error.

In this paper new efficient decision feedback equalization (DFE) schemes for channels with long and sparse impulse responses are proposed. It has been shown that under reasonable assumptions concerning the channel impulse response (CIR) coefficients, the feedforward (FF) and feedback (FB) filters may be also approximated by sparse filters. Either the sparsity of the CIR, or the sparsity of the DFE filters may be exploited to derive efficient implementations of the DFE. To this end, compressed sampling (CS) approaches, already successful in system identification settings, can significantly improve the performance of the non sparsity aware DFE. Building on basis pursuit and matching pursuit techniques new DFE schemes are proposed that exhibit considerable computational savings, increased performance properties and short training sequence requirements. To investigate the performance of the proposed schemes the restricted isometry property in the common DFE setup is also investigated.

Three new adaptive equalization algorithms for wireless systems operating over frequency selective MIMO channels are proposed. The problem of the MIMO DFE design is formulated as a set of linear equations with multiple right-hand sides (RHS) evolving in time. By applying an adaptive modified conjugate gradient algorithm, originally proposed for a single linear system, to the problem at hand, we arrive at an equalizer of performance identical to RLS, numerically robust, but of higher computational cost. To reduce its complexity, two updating strategies of the equalizer filters are derived based on Galerkin projection in time and space respectively. The two alternative schemes exhibit a complexity lower than RLS while offering slightly inferior convergence properties.