The integration of the Social Internet of Things (SIoT) with Wireless Sensor Networks (WSNs) significantly enhances the efficiency, scalability, and intelligence of distributed sensing systems. However, these networks often encounter severe resource constraints, limiting the control and management traffic that can be introduced. WSNs, typically composed of battery-powered edge sensor nodes with limited computational capabilities, memory, and communication bandwidth, face challenges in optimizing performance. In this work, we address the multi-objective optimization problem within the context of resource-constrained SIoT, aiming to reduce the energy consumption of edge nodes while simultaneously enhancing the quality of the received data based on channel conditions. We propose a Pareto Optimization framework to jointly optimize the compression factor and coding rate in a WSN scenario utilizing LoRa technology for communication. This framework explores the trade-offs between energy consumption and data reconstruction quality, leveraging Compressive Sensing (CS) for efficient data compression to alleviate the transmission load on edge nodes. Furthermore, we present a distributed optimization solution to minimize energy consumption while maximizing data quality, thereby reducing signaling and control overhead. This study contributes to the development of energy-efficient, scalable, and sustainable SIoT systems by providing a foundation for optimizing data transmission in LoRa networks.

Low-Power Wide-Area Network (LPWAN) technologies such as LoRa are key enablers of large-scale IoT monitoring systems, where long communication range and low energy consumption are essential. These advantages, however, come at the cost of strict throughput limitations, which significantly shape IoT system design. Compressive Sensing (CS) can mitigate this constraint by reducing transmitted data volumes, effectively trading communication load for additional processing at the sensing node and receiver. From an IoT perspective, this shift impacts node lifetime, hardware requirements, and overall network scalability. In this paper, we propose a data-driven optimization framework for LoRa-based IoT sensing systems employing CS. The approach jointly analyzes reconstruction quality and energy consumption through surrogate regression models that capture the interaction between physical-layer parameters and compression levels. This enables efficient multi-objective optimization via Pareto-front analysis and utopia-based selection. Results show that CS does not always dominate the quality-energy trade-off and that unified and stratified surrogate strategies identify closely aligned optimal operating points. Overall, the framework provides a practical and interpretable tool for the design of energy-efficient IoT sensing deployments.

The integration of Social Internet of Things (SIoT) paradigms with Wireless Sensor Networks (WSNs) offers significant improvements in the efficiency, scalability, and intelligence of distributed sensing systems. However, these networks are often subject to severe resource constraints, particularly at the edge, where sensor nodes are typically battery-powered and limited in computational power, memory, and communication bandwidth. Consequently, the introduction of control and management traffic must be carefully limited to avoid compromising network performance. In this work, we tackle the problem of multi-objective optimization in resource-constrained SIoT environments. Specifically, we aim to reduce the energy consumption of edge sensor nodes while improving the quality of the received data, taking into account the underlying channel conditions. To this end, we propose a distributed optimization strategy that minimizes energy consumption and maximizes data quality, while implicitly reducing the overhead associated with signaling and control messages. This framework explores the trade-offs between energy efficiency and data reconstruction accuracy, leveraging both Compressive Sensing (CS) to effectively reduce the transmission burden, and channel coding techniques to enhance data protection in LoRa-based WSNs.

The Widespread adoption of randomization in modern operating systems has introduced significant challenges for passive monitoring and user tracking in wireless environments. These challenges are further increased in large-scale environments covered by multiple access points, where associating transmission across zones becomes more complicated. This paper presents a frame association-based approach that enables cross-zone tracking and trajectory reconstruction using Wi-Fi probe request frames. The proposed method correlates transmissions across multiple access points by analyzing a combination of fingerprints. Results show that our approach effectively associates frames transmitted from the same origin, tracks devices across multiple zones, and provides insights into user movement and behavior, such as the type of transitions between zones and the reconstructed trajectory, all while preserving user privacy.

Accurate assessment of postural stability is crucial for monitoring patients with movement disorders, as it helps detect early signs of instability and prevent falls. Traditional methods, such as force platforms, are expensive, bulky, and limited to specialized laboratory settings, making them unsuitable for regular clinical screening or continuous home-based monitoring. In this work, we propose a deep learning approach to predict the Center of Pressure (CoP) trajectory using Inertial Measurement Unit (IMU) data from wearable smart glasses, offering a portable and cost-effective alternative. We use synchronized data from a force platform and a 9-axis IMU sensor to model the relationship between raw IMU signals and CoP force data (Force X and Y). The method involves windowing the IMU data, preprocessing it with low-pass filtering, and applying normalization. The dataset includes sequences from three standing tasks (eyes open, eyes closed, and free stance), captured at a frequency of 100 Hz. Experimental results show that the LSTM and BiLSTM models accurately predict CoP trajectories, achieving low Mean Absolute Error (MAE), Mean Squared Error (MSE), and high R2 values. While the TCN and GRU models face certain challenges in achieving the same level of performance as LSTM and BiLSTM, they present valuable insights and potential for future refinement. This approach has the potential to enable real-time, portable balance monitoring and early detection of postural instability, offering a scalable solution for clinical settings and home-based monitoring.

Enabled by the integration of AI and the Internet of Medical Things (IoMT), smart and remote monitoring systems are poised to play a pivotal role in the future of healthcare. Specifically, monitoring respiration is a critical component of this evolution, offering a straightforward yet effective method for assessing an individual's health status. In this paper, we introduce an innovative contactless approach to respiration monitoring that leverages Channel State Information (CSI) from Wi-Fi channels. By analyzing variations in the amplitude and phase of CSI data, we infer human respiration patterns. To validate our method, we conducted experiments with multiple subjects in indoor environments, assessing the system's capability to track the respiration cycle and determine breathing frequency. We also compared the performance of various AI algorithms in identifying accurate breath rates. Our findings indicate that the CSI-based system is a promising solution for respiration monitoring, achieving an average accuracy of approximately 84 % in estimating breathing frequency, thus paving the way for future studies to enhance the robustness of the proposed approach.

The Internet of Things (IoT) has garnered significant attention in recent years, with the integration of AI solutions and wireless sensing technologies enabling innovative approaches to context awareness and user location. Additionally, Channel State Information (CSI) from WiFi channels is emerging as a key component in next-generation wireless systems. In this work, we conduct a comprehensive analysis of the sensing capabilities of state-of-the-art CSI tools, namely the Intel 5300 and the ESP32 CSI tools, through extensive experimental tests in a dedicated testbed. The results offer valuable insights into CSI-based techniques, demonstrating their strong potential for activity detection and context aware applications.

This paper presents the design, development, and implementation of an Industrial IoT (IIoT) node aimed at monitoring vibrations on various types of structures, such as bridges, lightning rods, and large industrial machinery. The IoT node leverages a ESP32 microcontroller, an Inertial Measurement Unit (IMU), and radio communication technology for data transmission. The proposed solution is capable of gathering and transmitting real-time data from accelerometers, gyroscopes, and magnetometers via a LoRa interface. In this work we carry out a fine system calibration procedure, to assess the actual features of the IIoT node and provide a thorough experimental analysis of the performance of LoRa technology in complex urban environments performing extensive field tests. This work lays the foundation for the utilization of the IIoT node equipped with LoRa technology in both urban environments and industrial IoT frameworks, highlighting its adaptability and potential for wide-scale applications in these settings.

Mobile devices with enabled Wi-Fi continuously discover nearby access points by broadcasting management frames known as probe requests. This broadcasting leaks fingerprints that can be used to identify the presence and movement of flow in a given area. To protect users' privacy and location, probe requests use randomized MAC addresses generated according to the randomization protocols. In this paper, we analyze the behavior and characteristics of probe request frames across different devices from various vendors, operating systems, and features, as well as the influence of user-device interaction in different phases. In particular, we provide a detailed examination of the adoption of MAC address randomization techniques to highlight the strengths and weaknesses of recent randomization policies and to demonstrate that, despite the OS implementation, there is still potential to utilize open Wi-Fi in different services and development.

The convergence of Symbiotic IoT, AI foundational models, and 6G is ushering in a new era of intelligent connectivity, where networks, devices, and algorithms operate in close coordination to enable real-time, adaptive, and efficient systems. In the context of Structural Health Monitoring (SHM), this integrated vision provides a powerful framework to tackle challenges such as limited resources, harsh environments, and the need for timely, high-fidelity data. By enabling intelligent, collaborative processing across edge and network layers, it supports efficient data compression, transmission, and decision-making—ensuring robust and adaptive monitoring even in complex structural settings. In this work, we investigate the use of AI for data compression in an IoT-based SHM scenario. Specifically, we evaluate and compare the performance of four different Convolutional Autoencoders (CAEs) in compressing and reconstructing inertial signals collected from various structural systems, aiming to enable adaptive and context-aware processing directly at the edge. By testing across heterogeneous sources, we assess the generalizability and robustness of each CAE model, providing insights into the potential of deep learning-based compression techniques for SHM applications.

The use of multiple access protocols information in Internet of Things (IoT) environments has gained significant interest over the past decade, particularly for crowd behavior monitoring. Due to its high data rates and low infrastructure requirements, WiFi is considered one of the most promising wireless technologies for leveraging the explosive growth of transmitted data from mobile devices. However, with the introduction of MAC address randomization and the application of new randomization policies on assigning randomized sequence numbers (SNs) to transmitted frames in mobile devices to enhance privacy, traditional approaches to device identification face significant challenges. To tackle this issue, we conduct a comprehensive analysis at the multiple access level and propose CrowdWatch which is an innovative framework designed to enhance the understanding and utilization of MAC randomization dynamics. Additionally, we introduce a novel approach that leverages multiple device-specific features to accurately associate frames with randomized MAC addresses to their corresponding devices. By integrating multimodal fingerprints, the framework can effectively identify mobile devices and track their behavior. The presented approach ensures reliable detection even under the latest randomization policies. We examined the introduced framework through real-world experiments, and the findings prove that it is an effective solution for smart building management and occupancy estimation in dynamic environments.

In recent years, the increase in the elderly population has placed significant burdens on post-rehabilitation schemes, resulting in high logistical costs and considerable social impacts due to hospitalization or frequent visits. These challenges call for a transformation in the traditional approach to physical patient care, which can be achieved by leveraging the Internet of Medical Things (IoMT), particularly through the use of pervasive wearable sensors. When attached to patients during treatment or therapy, these sensors can provide valuable supplementary information to healthcare professionals. When it comes to adopting IoMT technologies, cost efficiency, portability, and generalization are key factors. Specifically, this study aims to enhance the cost-effectiveness and versatility of wearable eHealth monitoring architectures that utilize foot pressure-sensing hardware for the motor assessment of post-stroke and neurologically impaired patients. It leverages lower limb inertial measurement unit sensory information and machine learning to mitigate the reliance on foot pressure-sensing hardware. We demonstrate the potential of artificial intelligence (AI) in predicting fine-scale foot pressure using only inexpensive, off-the-shelf motion sensors. We propose a self-supervised, exercise-agnostic asynchronous foot pressure decoding model that does not require human annotation. The algorithm is thoroughly evaluated using appropriate performance metrics, and our experimental tests show promising results.

Drones are integral to various applications, out of which traffic surveillance is an important application. However, their operational efficiency is limited by battery life, which restricts their capacity for extended critical missions. Additionally, in remote or high-interference areas, the bandwidth for drone communication is often limited, leading to a decrease in the quality of images transmitted to the base station. This paper aims to address such challenges by having drones transmit video data in real-time at lower resolutions for traffic monitoring. This approach conserves energy and optimizes transmission. However, it adversely affects object detection accuracy at the base station due to compromised data quality. To address this issue, we incorporate Generative Adversarial Networks (GANs) to improve LR images, restoring their quality for precise object detection. Results indicate that the accuracy of traffic analytics achieved with GAN-enhanced images is comparable to that obtained with high-resolution data transmission. Consequently, our approach allows a fundamental trade-off among drone energy consumption, transmission time, flight time, and object detection accuracy, enabling robust detection performance while conserving energy and enhancing operational capabilities.

Contactless patient monitoring is one of the most trending topics in eHealth, due to the utmost importance of non-invasive tele-enabled biomedical systems in next-generation healthcare. In this connection, this work investigates Human Activity Recognition (HAR) using commodity 5GHz WiFi devices, exploiting Channel State Information (CSI) to distinguish among a set of actions performed by different people. The position and movements of the human body affect wireless signal reflections and, consequently, CSI. Data collected from wireless packets are organized into the CSI matrix, which describes the status of the link at each time instant. In this work we employ amplitude and phase information, related to the variations in CSI values, to classify human activities leveraging simple machine learning techniques. Different radio link modes are also compared to evaluate their impact on the classification performance. Experimental results show that CSI data is capable of providing very accurate results in classifying activities performed by different people, especially when considering phase-related information in a multiple-input-multiple-output (MIMO) configuration, thus making CSI-based HAR a promising solution for contactless patient monitoring.

Human balance is essential for everyday activities, from basic functions like standing and walking to more complex movements required in sports and work tasks. Optimal balance reduces the risk of falls, a major cause of injuries among the elderly and individuals with certain medical conditions. However, traditional force platforms can be prohibitively expensive for smaller clinics and individual practitioners, and patients with mobility issues may find it difficult to access locations equipped with such platforms. In light of these challenges, this work explores the effectiveness and versatility of using simpler hardware for pressure sensing. As an alternative to conventional force platforms, low-cost wearable sensors, such as Inertial Measurement Units (IMUs), have been explored. This research focuses on developing a simple yet effective balance evaluation system using smart-glasses embedded with an IMU sensor to replace traditional feet pressure sensing machines. Furthermore, we developed a custom dataset for the estimation of sway parameters by simultaneously collecting the IMU data and labels from the force platform for a set of 20 participants. We experimented with various Deep Learning (DL) models, leveraging the latest advancements in Machine Learning (ML), to estimate sway parameters such as sway path, sway area, and their ratio, typically measured by the gold standard force platform. When evaluated with appropriate performance criteria, the experimental results indicate that our proposed methodology performs robustly using only accelerometer data.

Structural Health Monitoring (SHM) is a process aimed at studying variations in the expected behavior of a structure in order to locate damage, material deterioration and other abnormalities. To this aim, SHM is usually performed continuously, thus generating large amounts of data, often by employing wired, expensive and proprietary systems. Introducing low-cost, low-energy consumption and low-rate IoT devices allows for cheaper and easier installations also in scenarios where computation and transmission resources are limited. Since many structural signals (e.g., vibrations) are sparse in the frequency domain, it is possible to apply well-known Compressive Sensing (CS) techniques to limit the amount of information to be transmitted. CS allows recovering a vector using a reduced amount of entries, thus being able to perform sub-Nyquist sampling. This paper shows the results obtained by applying CS to inertial signals coming from wireless IoT devices, developed as laboratory prototypes, applied to real structures (specifically, a bridge). Such findings are further expanded by discussing the efficiency of CS with respect to the number of used samples and its feasibility for IoT applications, from the transmission burden and energy consumption standpoints.

The eHealth systems are in great demand, particularly during times of outbreak like COVID-19, when there is a shortage of caregivers. The technological advancements, such as wearable wireless devices, the Internet of Things, and improved machine learning methods have made these systems more reliable. In modern times, these systems can play a vital role in post-rehabilitation journeys, which have significant social impact and high costs in traditional settings. Cost efficiency, portability, and generalization are key factors in adopting new technology. In this study, we investigate the potential for optimizing and simplifying hardware in order to increase the cost-effectiveness and versatility of post-stroke eHealth rehabilitation systems. It leverages the rich information available from Inertial Measurement Unit (IMU) sensors to compensate the need for foot pressure sensing. We present the first attempt to demonstrate the potential of machine learning, aided by affordable off the shelf motion sensing devices, for foot pressure analysis. Our proposed foot pressure decoding model is trained in an exercise-agnostic, self-supervised manner that eliminates the need for human annotation. The algorithm is evaluated using appropriate performance metrics, and our experimental results show very promising performance.

Compressive Sensing (CS) is a sampling technique which, provided the sparsity of the arrival domain and specific properties of the reconstruction matrix, allows rebuilding a vector starting from a significantly small subset of measures. This paves the way to a plethora of applications, ranging from specialized frameworks, such as medical imaging, to more general purposes, such as data compression. Among these, Structural Health Monitoring (SHM) is a primary and current topic, focused on analyzing structures to determine their residual lifespan and their health conditions (material degradation, damage localization, disaster prevention, etc.), In this regard, CS is able to provide accurate results, at the same time limiting the amount of data needed to propagate information between different end-points. Indeed, SHM usually deals with continuous flows of information originating from heterogeneous sensors and locations, often characterized by diverse computational power and signal coverage. In this paper we apply CS to signals coming from two different structures, i.e., a laboratory model and a bridge. Results show that CS is a viable way of reconstructing the considered signals by exploiting a subset of samples while still maintaining a high degree of precision, achieving an average normalized RMSE of 0.12.

Compressive Sensing (CS) is a sampling technique that challenges the traditional sampling scheme introduced by the Whittaker-Shannon theorem. Under certain conditions, a signal can be sampled at rates lower than the Nyquist rate, introducing a different kind of approach to signal handling, both in the acquisition and in the reconstruction phases. CS relies on the property of sparsity, the idea that a signal possesses an amount of information which is smaller than the amount of data required to store it. This paper employs the CS approach to inertial signals sensed by innovative IoT devices by showing applications of real-world infrastructure monitoring. Numerical results show that our approach is able to efficiently estimate the infrastructures modal frequencies with an innovative inertial IoT prototype by achieving a compression level around 20 times below the Nyquist rate.

Alongside many traditional as well as novel applications, in the latest years, drones have been widely adopted as remote sensing platforms for road traffic monitoring in urban areas and on highways. The problem of traffic monitoring on Region of Interest (RoI) based on drone imagery is a challenging task, especially when the surveillance drone is constantly moving. In this work, two specific subtasks have been addressed. The goal of the first stage is to predict the RoI in drone imagery of traffic scenes using deep-learning (DL)-based approaches instead of traditional image processing; in this connection, the goal of the second task is to perform vehicle detection on the selected RoI. To ensure diversity and robustness, drone images with different altitudes, positions, and view points have been considered. To achieve these goals, two custom aerial data sets for RoI extraction and detection were built by collecting aerial sequences from flying unmanned aerial vehicles (UAVs) and by transmitting them to the base station leveraging 5G technology. Two different ad hoc DL-based architectures have been designed for the RoI extraction task to maximize the accuracy and inference speed, respectively, and have been evaluated on two different data sets: 1) a custom-built data set and 2) a Massachusetts roads data set. Finally, the models providing the best performance have been combined to further improve the overall results. Experimental tests show that the proposed framework represents a promising solution for drone-based road traffic monitoring in critical areas, exploiting imagery from a variety of viewing angles and altitudes.

Drones and video processing have become a vital and integrated aspect of smart city development in several applications such as search and rescue, surveillance and delivery. Newly, camera-equipped drones are flown, and high-resolution photos and video data are relayed via a communication link to the base station. Apart from tackling the video processing issues in applications, such as object identification and tracking, energy consumption that may present a stumbling block in completing a successful drone flight for data collecting must be highlighted and considered. Drones have a limited amount of energy storage, which must be used to power the drone's movement, hovering, data collection, and communication. This study aims at building and testing a drone energy profile that estimates the total energy consumed and a maximum flying time of a drone in a traffic monitoring scenario. Additionally, the relationship between the video processing task and the drone energy profile is explored to determine the optimal strategy for flying the drones while maintaining the video processing task's performance. The conclusion of this evaluation and investigation can be conceived of as the test-case scenario to fly a drone for surveillance and monitoring applications to attain the optimum results.

Unmanned aerial vehicles (UAVs) are increasingly utilized in smart cities to perform traffic monitoring tasks such as multiple object detection and tracking. The task's criticality is dependent on the drones' dynamic altitude, movable camera, and various viewing angles. These challenges are addressed once the UAVs' collected data is received. However, parameters affecting drone data collecting flight operations must also be explored, including drone actual flight time, data collection time, and energy consumption profile. The drone flight time would depend upon the battery capacity and energy consumption profile, which comprises drone movement, data collection, and communication energies. Besides, data collection energy consumption is subjected to video quality, frame rates, and data compression. The installed battery in UAVs is of limited capacity and determining actual flight time, data collection time, and energy consumption profile based on the factors mentioned above is critical. This paper develops and examines a drone energy consumption profile and proposes a drone flight strategy in a surveillance scenario to correctly estimate the drone's actual flight time, data collection time, and the distance the drone could travel from its original location. The results of this analysis are presented as a test case for flying a drone to collect the traffic monitoring data.

From smart cities development perspective, road vehicle detection exploiting drone-based aerial imagery is a crucial part of traffic surveillance and monitoring systems where effective results are of utmost demand. A recent boom in the field of deep learning (DL) has provided remarkable development in the problem of vehicle detection. Aerial views pose more complexity with respect to the ground view but the rapid advancement in the field of DL, the volume of data, and hardware configuration has facilitated the realization of these intelligent detection systems effectively. In this article, a detailed performance evaluation of some of the main state-of-the-art DL-based object detection techniques has been carried out along with an experimental analysis of vehicle detection using the RetinaNet framework on the VisDrone-benchmark data set. The performance of the RetinaNet framework has been validated together with the results provided by the VisDrone team. Further experiments are then conducted to investigate the impact of various parameters. Finally, the selection of suitable models that can be practically implemented is also discussed based both on a qualitative and quantitative analysis.

According to data traffic forecast reported in Cisco (2018), the volume of data transported by Internet in 2021 will exceed the threshold of 3.0 ZB per year, generated by roughly 30 billions of devices. SATCOM’s capability to bring Internet connectivity in parts of the world that until few years ago was impossible to reach (e.g., in the middle of the ocean), the use of satellite platforms for M2M services has become more and more widespread, with an unprecedented increase over the last five years (see Herrero and De Gaudenzi, 2012 and Scaliseet al., 2013) and expected further diffusion in the near future. One of the most interesting applications, enabled by the Internet of Things (IoT), that has attracted significant attention in the M2M field is the personal vocal assistant (PVA). It is a software agent that can perform tasks or services for an individual based on commands or questions given through voice instructions. The main contribution of this article is a performance analysis, both theoretical and simulated, of satellite networks for PVA services over cruise ships. Namely, a PVA must provide feedback to the user in a short amount of time to guarantee a high Quality of Experience (QoE) Zhanget al., 2013. For this analysis, we tested different satellite providers and we come to a conclusion regarding their feasibility as a reference network for the PVS employment on a cruise ship. Future studies will investigate the proposed scenario by broadening the performance analysis through actual measurements over a real-world satellite channel to practically demonstrate the validity of our approach.

The leading trends in the framework of the Internet of Things are driving the research community to provide smart systems and solutions aimed at revolutionizing medical sciences and healthcare. One of the major opportunities offered by IoT lies in the ubiquitous connectivity, thus enabling smart services such as remote patient monitoring, in-home therapy/rehabilitation, and assisted living platforms. In this paper we present a prototype of wearable smart glasses able to monitor the Eye Blinks (EBs) through ElectroOculoGram (EOG) signal in a transparent way with respect to the final user. We propose a novel pre-filtering scheme to reduce EOG noise along with an analytical derivation of a matched filter to detect and count EBs. We have carried out an in-depth experimental campaign in order to validate the robustness of our approach with respect to the main solutions available in the literature. Furthermore, we have compared the performances obtained with out wearable prototype versus the results achievable with professional medical equipments. Results show that our solution is able to achieve very high accuracy in EB detection, obtaining comparable performance with respect to professional medical desktop equipment, with the additional benefit of portability, comfort and easiness of use for the patients.

Given the extreme diffusion of Alzheimer's disease (AD) and Parkinson's disease (PD), the necessity for a solution to early detect neurological symptoms of such diseases strongly arose. According to the medical literature, such early detection can be achieved by exploiting the correlation between PD and AD and some external symptoms: the Essential Tremor (ET) and the number of Eye Blinks (EBs). In this paper we present a prototype of sensored glasses able to detect the presence of ET of the head and to count the number of EBs at the same time. To the best of authors' knowledge this is the first attempt to monitor such external symptoms with a transparent and wearable device without any a-priori training. Numerical results prove the reliability of the proposed approach: the proposed algorithms are able to i) correctly recognize the ET with an overall accuracy above 97% and ii) count the number of EBs with a Root Mean Square Error (RMSE) around 0.4.

In the latest years, the popularity of commercial drones has grown rapidly due to their cheaper costs and great availability on the market. The great diffusion of remotely piloted devices unfortunately leads to several security and safety concerns that need to be tackled. In this paper, we consider a fingerprint-based drone detection approach relying on the analysis of WiFi traffic features to identify the presence of unauthorized devices. In particular, we study the statistical distribution of the features composing the fingerprint vector, and we propose an adversarial approach to drone detection in order to invalidate the reliability of the surveillance system, by introducing fake ad-hoc traffic features. Results show that our novel approach is able to deceive the drone detection system through the introduction of flows belonging to arbitrary traffic classes. Also, the proposed adversarial method provides the expected significant impact on the performance of the system, actually reducing the recognition accuracy to about 50%.

In the latest years we have been witnessing an evolution of the technological framework thanks to the Internet of Things paradigm, which is enabling innovative smart services for many different applications. The healthcare system is among the main scenarios that are benefiting from this new trend, thus giving rise to the concept of eHealth. In this framework we propose SmartPANTS, an IoT-based wireless system specifically designed for the rehabilitation of lower limbs. Our system is conceived as a prototype of a medical platform to be employed during the physical therapy for patients recovering from a brain stroke condition. SmartPANTS includes a signal processing and machine learning algorithm able to automatically recognize the type of exercise the patient is performing, and to provide real-time feedback on the execution. Moreover, experimental tests show that our platform is able to estimate the execution time of the different exercises, providing values very close to real ones. Performance results show that the SmartPANTS system is able to correctly identify the type of exercise currently being performed with an accuracy of about 99%.

The growth of Internet traffic especially with respect to mobile broadband services is one of the main motivations towards the integration of satellite and 5G networks. A key building block for the overall network orchestration will be the efficient definition and management of network slices, which eventually translates into efficient allocation of physical and computing resources. This paper explores the resource allocation problem in the case of railway communications building on satellite and 5G connectivity. Some exemplary candidates solutions have been elaborated, whose validation through simulation campaigns has shed some lights about the main factors impacting on the overall system performance and the most effective resource allocation solutions.

Information related to recognizing the place in which a user is has become crucial to deliver efficient and tailored Location-Based Services (LBSs). Though plenty of solutions exist for recognizing indoor places, almost no idea is present in the literature for recognizing big outdoor places without the GPS employment. Though these solutions assure great accuracy they also have a strong request in terms of energy necessary to achieve such result. This paper proposes a Place of Interest (POI) recognition algorithm called Enhanced-LRACI. It is an evolution of LRACI (Location Recognition Algorithm for automatic CheckIn applications), a former work reported in [1]. E-LRACI aims at recognizing big outdoor places only by using WiFi Access Points (APs) over mobile devices (smartphones). The main contributions of this work are: i) it solves the problem of big outdoor POI recognition without using GPS by leveraging on the concept of spot, ii) it proposes a novel fingerprint (FP) algorithm and iii) it outperforms the results obtained by other reference works in terms of recognition accuracy. Performance investigation reported in this paper, carried out on real data acquired over mobile devices (Android smartphones), shows that our proposal is able to correctly recognize big outdoor POIs in 95% of the cases whereas other state of the art papers do not exceed 89%.

The increasing popularity of low-cost aerial vehicles that can be remotely piloted by amateurs, is giving rise to a number of issues related to public safety and privacy. At the same time, the need for surveillance methods able to detect the presence of unauthorized drones meets the ubiquitous connectivity and pervasive technology typical of the big data era. This dualism drives the research community towards the design of monitoring systems able to integrate machine learning methods and data mining techniques in this evolving environment. In this framework, we consider a WiFi based approach aimed at detecting nearby unmanned aerial vehicles, by performing statistical fingerprint analysis on wireless traffic. We study the inherent vulnerabilities of the considered method through real-life experimental tests by setting up specific attack scenarios, and we devise and test a solution in order to improve the efficiency of the proposed technique in the presence of malicious countermeasures. Results show that the proposed improved detection technique is indeed robust to stealth attacks, and it is able to achieve very good recognition performance in different real-life testing scenarios.

Indoor positioning systems have gained a lot of attention in the last few years. With the introduction of the Internet of Things (IoT) paradigm, the knowledge of user's location has become crucial information to deliver the efficient and tailored Location Based Services (LBS), especially in indoor environments. In this paper we propose an AP (Access Point)-centred indoor positioning system that overcomes common limitations presented in conventional positioning systems, such as an excessive involvement of Mobile Devices (MDs). Our work merges ideas originally proposed in [1] and [2] to build an efficient, accurate and smart Probabilistic-FingerPrint (P-FP) algorithm that avoids the MD involvement and considers the signal strength measurements as a random variable in the positioning process. Numerical results, obtained in a real-world deployment, show better performance on positioning accuracy, energy consumption and latency with respect to the MD-based architecture.

Nowadays, Internet of Things (IoT) devices can collect a large amount of data and infer the context they are operating in. One of the most pervasive IoT device is the smartphone. Among the plethora of sensors that such device has, microphone is probably the most versatile. It can be used to infer information about the context by acquiring the environmental sound. In this paper, we propose an active ultrasonic-based method able to sense context information. The approach is based on the emission of periodic ultrasonic impulses, called pings, whose echoes are continuously acquired to extract a proper set of features. Using a classifier the context information is then retrieved. Two applications are presented: 1) an indoor/outdoor detector and 2) an earphone wearing state detector. In both cases a smartphone is the employed device. The former senses if the smartphone is in an indoor or outdoor environment while the latter detects if a user is wearing or not his earphones. The obtained results are encouraging for both the solutions that have been stressed in different working conditions and employed within proper application frameworks.

The treatments for brain stroke are strongly time- dependent. The medical literature highlights the need of a quick diagnosis in order to guarantee the most effective therapy. An important target for strokes is trying to achieve a Door-to-Needle (DTN) time of less than 60 minutes, which is called Golden Hour [1]. This paper proposes a mobile Smart Helmet (SH) thought to be worn by a patient when the first aid medical team arrives and the aim is to efficiently recognize and detect a brain stroke, on site. While similar solutions in the literature employ the (usually computationally heavy) electromagnetic field inversion problem and image analysis, the proposal of this paper is an NN-based SH. It uses signal analysis to recognize the presence of a stroke with a limited computational burden. In the reported preliminary experiments, carried out via simulations, we have employed a MultiLayer Perceptron (MLP) model that implements a 4-layer NN. Numerical results show that proposed signal analysis, applied to a single brain model, is able to efficiently detect the stroke presence with an accuracy around 90%.

Location Fingerprinting (LF) is a promising localization technique that enables many commercial and emergency Location-Based Services (LBS). The idea of this paper is two- folded. First, a Gaussian Process (GP) is used during the training (offline) phase of an indoor positioning algorithm to generate the fingerprint database, reducing the expensive labor of acquiring and maintaining the fingerprint database significantly. Furthermore, during the positioning (online) phase, a Smart algorithm (already proposed in [1]) is used for reducing the computation effort for positioning calculation. We call our idea Smart2since it enhances the advantages of the base Smart approach. Specifically, Smart2reduces the labor during the offline phase by trading it with a small positioning error and, at the same time, it limits the energy consumption in the online phase without incurring in any additional accuracy detriment.

Rehabilitation and elderly monitoring for active aging can benefit from Internet of Things (IoT) capabilities in particular for in-home treatments. In this paper, we consider two functions useful for such treatments: 1) activity recognition (AR) and 2) movement recognition (MR). The former is aimed at detecting if a patient is idle, still, walking, running, going up/down the stairs, or cycling; the latter individuates specific movements often required for physical rehabilitation, such as arm circles, arm presses, arm twist, curls, seaweed, and shoulder rolls. Smartphones are the reference platforms being equipped with an accelerometer sensor and elements of the IoT. The work surveys and compares accelerometer signals classification methods to enable IoT for the aforementioned functions. The considered methods are support vector machines (SVMs), decision trees, and dynamic time warping. A comparison of the methods has been proposed to highlight their performance: all the techniques have good recognition accuracies and, among them, the SVM-based approaches show an accuracy above 90% in the case of AR and above 99% in the case of MR.

Nowadays, the widespread use of Mobile Devices (MDs) open the door to exploit the presence of multiple nodes to accomplish collaborative tasks. In this paper, a speaker recognition system for MDs based on a multiple-observations approach is presented. We propose different fusion and clustering algorithms aimed at efficiently exploiting signals coming from multiple sensors. Numerical results show that in most cases our multiple-observations approach is able to significantly improve the performance of a single- receiver approach.

This paper focuses on countering data hiding in images. Specifically, we refer to the active warden problem, a process which aims at disrupting the covert communications that are possibly taking place within host media. Although this has been extensively studied in the literature, typical state of the art approaches treat the hidden message as a noise-like component, proposing de- noising techniques, lossy compression or noise addition as a countermeasure. In this paper we present novel approach called Lesser Components Distortion (LCD) that aims at disrupting the covert communication while minimizing distortion on the host medium. Although stemming from insights gained from notable works in the spread spectrum steganography field, we prove through extensive testing that it shows remarkable effectiveness against steganographic techniques in general, compared to common attacks.

In-home physical therapy is one of the best options for many individuals and families thanks to its convenience and because it makes possible to receive professional care in the comfort of your own home. To enable this therapeutic approach, this paper proposes the employment of a smartphone-centric platform for in-home rehabilitation. The platform helps physicians to monitor the patients remotely so avoiding hospitalization therapies that can be stressful. In more detail, the work is focused on the Movement Recognition (MR) functionality of the aforementioned platform. It compares algorithms, which process the signal provided by the embedded accelerometer sensor of the smartphone, able to recognize if a patient had performed the movements requested by the physicians. The provided performance comparison of different MR techniques shows that Support Vector Machine-based approaches have very good accuracy (up to 99.3%), thus making the in-home physical therapy reliable.

Recently the problem of interconnecting Internet of Things nodes when they are dispersed over wide geographical areas has been highlighted: this is the Internet of Remote Things (IoRT). It could be able to play a key role in disaster recovery scenarios (e.g., earthquakes, flash floods, terrorist attacks, etc.), where the presence of a communication infrastructure enabling video transmissions from the emergency areas is crucial. Local communications can be provided by incident area networks, while satellite communication systems can bridge the gap toward external networks. In this context, we propose a cooperative resource allocation mechanism, which exploits feedback from links conditions and knowledge of the adaptive coding, allowing a higher quality and more fair experience for all nodes of the IoRT transmitting videos. We implemented each element of the mentioned scenario over an emulation platform to prove the concept and show the obtained benefits.

The exponential increase in the number and types of mobile devices, along with their ever-growing sets of capabilities, have enabled the development of new architectures that aim to harness such heterogeneity. Transient Clouds (TCs) are examples of mobile clouds which are created on-the-fly by the devices present in an environment to share their physical resources (e.g., CPU, memory, network) and would disappear as the nodes leave the network. They enable a device to go beyond its own physical limitations through utilizing the capabilities offered by nearby devices over an ad-hoc network. In this paper we present a Transient Context-Aware Cloud (TCAC) in which the nodes of the network care more about providing/learning higher level functionalities rather than lower level capabilities. We make the case for such an architecture in scenarios where it is not feasible for all the nodes to compute the context due to privacy, energy, and delay constraints rather than an unreachable network.We present a prototype implementation of our architecture over Android smartphones connected via WiFi along with the performance metrics (power/energy consumption and accuracy)to show the benefits of context awareness in TCs.

This paper considers video transmissions through Smartphones in scenarios of emergency networks where satellite links are employed. Unfortunately, satellite channels may be affected by attenuation and fading, causing significant and frequent variations of the link quality. For this reason, static compression, coding and resource allocation are not optimal solutions to guarantee a satisfactory level of Quality of Experience: we propose to apply two techniques: i) a dynamic transmission rate allocation able to assign resources to the links according to their conditions; ii) an application layer joint coder for video transmission capable of adaptively compressing and protecting the transmitted video frames so recovering losses. The performance investigation of the proposal has been carried out through emulation and the obtained results are satisfactory and open the doors to future development of the proposal.

In this paper, an Android SPEech proCessing plaTform as smaRtphone Application (SPECTRA) is presented. Such application, developed by the authors, has three main functions: i) Gender Recognition (GR), ii) Speaker Recognition (SR) and iii) Language Recognition (LR). All these recognition functions are performed simultaneously by using unsupervised Support Vector Machine (SVM) classifiers. An innovative point of this paper lies in the automatic re-training of the employed SVMs which are able to dynamically update themselves when a (new) audio from a (new) speaker is provided. This allow to build more robust classifiers, which results in better recognition performances. In terms of accuracy, the GR reaches about 98% of correct classifications, SR performs around 80% while LR shows an accuracy of about 74%.

Context information is fundamental for mobile application. A system able to detect Indoor/Outdoor state switching can give useful information to upper-level applications permitting to improve their performance or reduce the computational load and consequently the lifetime of the smartphone. A localization application may exploit the Indoor/Outdoor information to smartly decide if using GPS (that performs well outdoors but poorly indoors) or other localization methods. In this paper we present an ultrasonic-signal-based Indoor/Outdoor detector for smartphones. The phone plays an ultrasonic ping by using its in-built speakers and records the echoes by using its microphone, therefore, no specific hardware is required to be added to the smartphone. The proposed detector shows good performance in terms of accuracy and latency.

No abstract available.

Often copyrighted multimedia files are uploaded and shared online. To avoid the unregulated spread of such material many copy detectors have been developed in order to deny the possibility to upload, and consequently make available, copies of copyrighted contents. A widely referenced fingerprint method for content-based audio identification is the Philips Robust Hash (PRH) . This paper introduces a simple but effective attack technique capable to defeat a PRH fingerprint-based audio copy detector without significantly affecting the signal quality. It is a heuristic method that adds a suitable distortion to the original audio signal, so that the modified signal is not detected as a copy of the original one but is perceptively very similar to it. The quality of the modified signal has been evaluated in terms of a distortion measure based on a mathematical model of the human auditory system and of the Peak Signal-to-Noise Ratio (PSNR). The attack method has shown a promising success rate.

Recently, people have been interested in sharing what they are watching on TV, allowing the development of Social TV Applications often based on mobile devices. In this context, this paper proposes IRTR (Improved Real-Time TV-channel Recognition): a new method aimed at recognizing in real time (live) what people are watching on TV without any active user interaction. IRTR uses the audio signal of the TV program recorded by smartphones and is performed through two steps: i) fingerprint extraction and ii) TV channel real-time identification. Step i) is based on the computation of the Audio Fingerprint (AF). The AF computation has been taken from the literature and has been improved in terms of power consumption and computation speed to make the smartphone implementation feasible by using an ad hoc cost function aimed at selecting the best set of AF parameters. Step ii) is aimed at deciding the TV channel the user is watching. This step is performed using a likelihood estimation algorithm proposed in this paper. The consumed power, computation and response time, and correct decision rate of IRTR, evaluated through experimental measures, show very satisfying results such as a correct decision rate of about 95%, about 2s of computation time, and above 90% power saving with respect to the literature.

Telemedicine applications provide healthcare services through communications technologies overcoming the geographical separation between patients and caregivers. These services can be provided via wireless devices, such as smart-phones with dedicated applications. An interesting application concerns the so-called situation awareness algorithms and, in particular, the Activity Recognition (AR) aimed at tracking the physical activity (or movements) of patients that need a constant monitoring of their medical conditions. This work takes as reference an architecture applicable, but not limited to, patients suffering from Heart Failure (HF) and presents a performance comparison between AR approaches based on the accelerometer signal captured through the patients' smartphones. In more detail, the considered AR techniques apply two different classifiers used to decide the patients movements: a J48 decision tree and a Support Vector Machine (SVM). For each classifier, three different features sets, characterizing the accelerometer signal, have been employed. The performance are evaluated both in terms of accuracy-related metrics and time needed by each classifiers to perform the decision. The results show that SVM provides the best accuracy while the J48 requires less classification time.

The Vertical Handover and the related Network Selection process play a fundamental role in supporting reliable communications over MANETs. The goal of the Network Selection is to determine the Radio Access Network (RAN) that a Mobile Node (MN) has to use among several available RANs. This decision process finds out the RAN that fits the MN requirements and has to provide the decision as rapidly as possible. A computationally heavy Network Selection algorithm can impact the whole handover process because it waits until the selection is carried out. This waste of time can have negative consequences in the quality of communications. The main contribution of this paper is the definition of a new Network Selection algorithm based on a different formulation of the Technique for Order of Preference by Similarity to Ideal Solution (TOPSIS) method found in the literature, called Dynamic-TOPSIS (D-TOPSIS). It is aimed at performing the same selection of the TOPSIS but with a significant reduction of necessary computational load. Finally, the numerical results highlight the execution time reduction obtained by applying the D-TOPSIS with respect to the standard algorithm, and other Network Selection techniques usually applied within Vertical Handover Processes.

The deployment of terrestrial/satellite networks, plays a crucial role for for risk and emergency management. In this context, efficient solutions for heterogeneous and mobile networks, including satellite portions that allows wide coverages, represents a key issue. In the mentioned scenario, transmitting video with portable devices (such as smartphone), over terrestrial/satellite networks, to a Remote Monitoring Host (RMH) may support emergency and rescue operations after crisis situations. Unfortunately heterogeneity often implies impairments such as packet losses, due to errors and congestion, which negatively affect the video quality. We present an application layer joint coding algorithm for video transmission, that adaptively applies video compression and channel coding at the application layer, on the basis of the overall network condition estimated in terms of maximum allowable throughput of the network and quality (packet cancellations or lossiness). A deep performance investigation, carried out with real implementation of the algorithm, compares the joint coding against fixed schemes and shows that the joint approach adapt the video transmission to terrestrial/satellite emergency networks so allowing a more efficient resource exploitation.

The paper deals with the classical problem of the resource allocation in geostationary satellite scenarios where fading may negatively impact the communications. In more detail, we consider the case in which different performance metrics, such as Packet Loss and Power, are simultaneously taken into account. Starting from the wide literature in the field, this work tries to categorize the available and well-known resource allocation approaches into three groups: i) the capacity maximization with constrained power group, ii) the power minimization with constrained capacity group and iii) the multi-objective programming group. For each group a general mathematical formulation has been proposed and the main differences among the groups have been highlighted. The performance of the three listed groups are compared through simulations where two Earth Stations, afflicted by different fading conditions, have been considered. The comparison takes into account all the mentioned metrics (i.e., loss and power) and the computational complexity of the allocation approaches.

This paper considers IRTR (Improved Real-Time TV-channel Recognition), a new method aimed at recognizing in real-time (live) what people is watching on TV, similarly to the action performed by Audience investigations, but without any TV user active interaction. IRTR uses only the audio signal of the TV program recorded through smartphones and is independent of the specific smartphone technology. It is performed through two main steps: i) fingerprint extraction and ii) TV channel real-time identification. This paper proposes a likelihood estimation-based algorithm aimed at performing the second step. The computational time of the proposed approach has been evaluated through real measures and shows really satisfying results.

This paper proposes a new location recognition algorithm for automatic check-in applications (LRACI), suited to be implemented within Smartphones, integrated in the Cloud platform and representing a service for Cloud end users. The algorithm, the performance of which is independent of the employed device, uses both global and hybrid positioning systems (GPS/HPS) and, in an opportunistic way, the presence of Wi-Fi access points (APs), through a new definition of Wi-Fi FingerPrint (FP), which is proposed in this paper. This FP definition considers the order relation among the received signal strength (RSS) rather than the absolute values. This is one of the main contributions of this paper. LRACI is designed to be employed where traditional approaches, usually based only on GPS/HPS, fail, and is aimed at finding user location, with a room-level resolution, in order to estimate the overall time spent in the location, called Permanence, instead of the simple presence. LRACI allows automatic check-in in a given location only if the users' Permanence is larger than a minimum amount of time, called Stay Length (SL), and may be exploited in the Cloud. For example, if many people check-in in a particular location (e.g., a supermarket or a post office), it means that the location is crowded. Using LRACI-based data, collected by smartphones in the Cloud and made available in the Cloud itself, end users can manage their daily activities (e.g., buying food or paying a bill) in a more efficient way. The proposal, practically implemented over Android operating system-based Smartphones, has been extensively tested. Experimental results have shown a location recognition accuracy of about 90%, opening the door to real LRACI employments. In this sense, a preliminary study of its application in the Cloud, obtained through simulation, has been provided to highlight the advantages of the LRACI features.

The task of a capacity allocation policy is to determine the optimal quantity of capacity that has to be shared among the transmitting entities. In this work the allocation problem is modelled by the Multi Objective Programming (MOP) theory. In particular, an allocation criterion based on the Lp-problem is proposed to find out a capacity allocation, among Earth Stations, representative of a compromise if Packet Loss Probability and Transmitted Power are taken into account as performance metrics. The paper also discusses the existence of a capacity allocation, called Capacity Bound, on which the performance converges independently of the overall capacity available CTOT. A performance analysis, carried out through simulations and under different satellite channel conditions, is finally proposed to investigate the allocation criterion performance and to show the Capacity Bound existence.

Recent multimedia and location-based services (LBSs) employ information about location, orientation and context of a mobile device. Moreover, the wide spread adoption of Smartphones, usually equipped with powerful processors, accelerometers, compasses and Global and Hybrid Positioning Systems (GPS/HPSs) receivers, has favored the increasing of location- and context-based services over the last years. In this work a novel Location Recognition approach aimed at supporting location aware services, in particular Check-In services, is presented. The proposed method uses both the GPS/HPS positioning information and, in an opportunistic way, the presence of Wi-Fi Access Points (APs). The method is based on the concept of radio FingerPrint (FP) whose definition proposed in this paper, to the best of authors' knowledge, has never been applied previously. The method is suited to be employed where traditional approaches - usually based only on GPS/HPS - fail, as in case of indoor or dense urban environments. Finally, the proposal, practically realized over different Android OS Smartphones, has been tested in terms of performance. The experimental results are very satisfying and open the doors to a wide employment of the method.

In this paper a method to realize the joint application layer coding for image transmission over deep space channels has been presented. In more technical detail, both image compression, based on algorithms such as JPEG2000 and CCSDS, and encoding techniques, such as LDPC codes, to protect the sent images are simultaneously applied by the proposed mechanism. It acts on the bases of the deep space channel conditions, in terms of Bit Error Rate, and it is based on the Multi-Attribute Decision Making theory. In practice, the proposal is aimed at protecting the essential informative contents of images sent through a deep space network and, at the same time, allows minimizing the load offered (the total amount of data to transmit) by the overall application layer coding process to the deep space network. The presented mechanism has been tested through simulations. The obtained results show the effectiveness of the proposal and open the door to further developments of the method in real systems.

The development of the new third generation cellular system, the UMTS, makes an extensive use of microcells and picocells other than macrocells: picocells are especially used to offer indoor radio coverage. Usually, indoor coverage design is modeled by disregarding the outdoor signal and by analyzing only micro and picocells placed indoor. This paper investigates the importance of the signal penetrating through windows and the measurements performed highlight how to decouple outdoor and indoor effects by modeling windows as equivalent radiating antennas. This solutions can be used as a precious support to indoor cellular design.

In this paper, we describe a method for fast determination of distortion-based optimal unequal error protection (UEP) of bitstreams generated by embedded image coders and transmitted over memoryless noisy channels. The UEP problem is reduced to the more general problem of finding a path in a graph, where each path of the graph represents a possible protection policy, with the objective of selecting the best path being that one inducing minimal distortion. The problem is combinatorially complex and excludes a brute force approach. The solution is provided by applying heuristic information from the problem domain to reduce search complexity. In particular, we use graph search procedure A suggested by Hart , well known in the field of artificial intelligence, to avoid exhaustive search. Numerical results show that this technique outperforms the method presented by Hamzaoui , in terms of Mean Square Error (MSE) distortion and computational complexity. After testing our solution using analytical models of the operational distortion curves proposed by Charfi , we implement a transmission architecture that, using the actual distortion values generated by a real embedded coder, computes the optimal protection policy for the considered image, protects the packets, and transmits them over a channel.

A method for the fast determination of distortion-based optimal unequal error protection (UEP) of bitstreams generated by embedded image coders is described. The UEP problem is reduced to the more general problem of finding a path in a graph, where each path of the graph represents a possible protection policy and the best path is that one inducing minimal distortion. The problem is combinatorially complex and excludes a brute force approach. The solution is provided by applying heuristic information from the problem domain to reduce search complexity. In particular, we use graph search procedure A, well known in the field of artificial intelligence, to avoid exhaustive search. Numerical results show that this technique outperforms the methods presented in [1], [2], in terms of mean square error (MSE) distortion and computational complexity.

The fast growth of cellular coverage demand for multimedia services based on wireless communication requires an increasing use of microcells to guarantee the satisfactory connectivity. In urban environments, especially in old town centres, the application of this solution is heavily limited by difficulties concerning coverage prediction and planning. The purpose of this paper is the investigation, development and testing of an analytical propagation model based on 2D ray-tracing approach for path loss prediction. Formulas are derived taking into account multiple reflections along street walls as well as diffractions at street corners, while ground reflections are neglected. Validation of the model has been carried out by using a data set collected during a measurement campaign in Genoa's old town centre.

The recent growth in personal wireless communication devices being used for image transmission, have poised a challenge to protect this data against loss over mobile radio channels. The paper addresses this problem investigating error protection schemes in the context of JPEG2000 compressed imagery. More particularly, the results reported in this paper provide coding guidelines concerning the selection of appropriate turbo code parameters along with optimized JPEG2000 error resilience tools.

This paper proposes a novel approach to energy efficient data transmission based on radio link estimation and adaptive channel coding. A channel protection algorithm is proposed to achieve efficient transmission (in terms of bit error rate), while reducing as much as possible power consumption and transmission bandwidth. The algorithm has been implemented and integrated within an experimental set-up based on a narrowband "ad hoc" network. It has been evaluated through field trials carried out in a hilly region with terrain characteristics that reproduce a variety of situations. The adopted simplified propagation model is based on the well-known knife-edge approximation while the radio link budget is computed by the mobile terminals thanks to 3D map information locally available. The level of channel protection is adaptively matched to the estimated radio link by tuning the transmission parameters and by configuring the Reed-Solomon protection code. Theoretical background and preliminary experimental results are presented in detail.

Non-verbal behaviors have a key role in making a virtual character appear life-like. We describe an extensible system for the specification, control and real-time generation of facial expressions and gestures. The system approximates in a MPEG-4 based virtual character the wide expressive range, dynamism (an expression's meaning significantly depends on its temporal evolution) and variability (an emotion is never expressed exactly in the same way by different people, and even by the same person at different times), typical of human nonverbal behavior. The MPEG-4 standard only allows high-level control of 6 basic emotions and does not explicitly support the description of an expression temporal evolution. Our approach has been that of creating a hierarchical model of expressiveness; expressions are defined in term of parameterized functions controlling low-level animation parameters trajectories (by means of an XML-based expression definition markup language). The real-time generation of those expressions is performed by an expression synthesis engine. The system allows to effectively modulate expressivity both at design-time (the developer tweaks the parameters to give the character a given expressive style), and at run-time (the engine automatically changes the way in which an expression is performed each time), producing controllable, but non-deterministic, behavior patterns, a key factor for enhancing believability.

The activity of the IST European Research Project INTERFACE is oriented to the development of advanced interfaces based on an interaction engine capable of interpreting natural voice or gestures and answering through artificial voice and animation feed-backs. Classic technologies of voice recognition and synthesis have been integrated with innovative algorithms for emotional speech modeling. Video analysis techniques have been developed for interpreting human gestures capable of being reproduced artificially through three-dimensional virtual animated faces and bodies. The INTERFACE objective is that of integrating all these analysis/synthesis techniques into a single piece of technology capable of accepting queries through talk and facial expressions, as we are used to do in everyday life with relatives, friends and colleagues, and of providing answers to the user through virtual actors capable of artificially reproducing human voice and gestures. The INTERFACE achievements have been recently demonstrated with live demos during international exhibitions and conferences.

An efficient algorithm is proposed to exploit the facial animation parameter (FAP) interpolation modality specified by the MPEG-4 standard in order to allow very low bit-rate transmission of the animation parameters. The proposed algorithm is based on a comprehensive analysis of the cross-correlation properties that characterize FAPs, which is here reported and discussed extensively. Based on this analysis, a subset of ten almost independent FAPs has been selected from the full set of 66 low-level FAPs to be transmitted and used at the decoder to interpolate the remaining ones. The performance achievable through the proposed algorithm have been evaluated objectively by means of conventional PSNR measures and compared to an alternative solution based on the increase of the quantization scale factor used for FAP encoding. The subjective evaluation and comparison of the results has also been made possible by uploading mpg movies on a freely accessible Web site. Experimental results demonstrate that the proposed FAP interpolation algorithm allows efficient parameter encoding at around 70 bits/frame or, in other words, at less than 2 kbits/s for smooth synthetic video at 25 frames/s.

We propose a method for implementing a high-level interface for the synthesis and animation of animated virtual faces that is in full compliance with MPEG-4 specifications. This method allows us to implement the simple facial object profile and part of the calibration facial object profile. In fact, starting from a facial wireframe and from a set of configuration files, the developed system is capable of automatically generating the animation rules suited for model animation driven by a stream of facial animation parameters. If the calibration parameters (feature points and texture) are available, the system is able to exploit this information for suitably modifying the geometry of the wireframe and for performing its animation by means of calibrated rules computed ex novo on the adapted somatics of the model. Evidence of the achievable performance is reported at the end of this paper by means of figures showing the capability of the system to reshape its geometry according to the decoded MPEG-4 facial calibration parameters and its effectiveness in performing facial expressions.

Advances in joint acoustical/visual analysis for model-based lip motion synthesis is presented. The 2D lip motion field is modeled as a linear combination of a low dimensional motion basis computed through principal component analysis (PCA). The vector of PCA coefficients is expressed as a function of a limited set of articulatory parameters which describe the external appearance of the mouth. The acoustical processing estimates these articulatory parameters from the direct analysis of the speech waveform based on a neural processing stage, i.e., through a bank of time delay neural networks. The achieved results have been subjectively evaluated by visualizing the estimated motion on a wire-frame mouth template presented in synchronization with speech. The experiments carried out so far deal with single-speaker trained TDNNs and with single-speaker PCA, but suitable algorithms for generalizing the techniques are currently under investigation.

A new technology is proposed for audio-video synchronization in multimedia applications where talking human faces, either natural or synthetic, are employed for interpersonal communication services, home gaming, advanced multimodal interfaces, interactive entertainment, or in movie production. Facial sequences, in fact, represent an acoustic-visual source characterized by two strongly correlated components: a talking face and the associated speech, whose synchronous presentation must be guaranteed in any multimedia application. Therefore, the exact timing for displaying a video frame or for generating a synthetic facial image has to be supervised by some form of speech analysis performed either as preprocessing before encoding or as postprocessing before presentation. Experimental results are reported on the use of time-delay neural networks (TDNN) for the direct estimation of the visible articulation of the mouth starting from the coherent analysis of acoustic speech. The architectural solution of employing a bank of independent single-output TDNNs has been compared to the alternative solution of using only a single multi-output TDNN. Similarly, two different learning procedures have been applied and compared for training the TDNN, the first based on the classic mean square error (MSE) and the second based on a measure of cross-correlation (CC). The superiority of the system based on multiple single-output TDNNs has been proved as well as the improvements, both in terms of convergence speed and estimation fidelity, achievable through the learning algorithm based on cross-correlation.

The research activity in natural/synthetic image processing and representation reported in this paper, initiated under the Esprit project HUMANOID and currently continued under the ACTS project VIDAS, concerns the application of virtual reality methodologies to interpersonal audio/video communication. The 3D videophone scene is modeled in video (the talker's face) and in audio (the talker's speech) so that natural data can be efficiently mixed with synthetic data and adapted onto deformable parameterized structures. Robust image analysis/synthesis tools are necessary to extract the visual primitives associated to the talker's face and to adapt them onto suitable modeling structures (wire-frames). Image/speech analysis performed at the transmitter provides suitable audio/video parameters which are encoded and used at the receiver to synthesize the corresponding facial expressions together with synchronized lip movements.

We approach the problem of designing the VQ codebook from a different point of view: after having evaluated a priori the effects produced by any possible single vector redistribution among clusters, we choose and apply the least distortion one. This way of looking at the problem with "new eyes", though quite immediate and simple in its formulation, provides powerful tools for devising a variety of new algorithms and procedures. The intrinsic property of the algorithm, assuring fast convergence in terms of number of iterations, has been fruitfully combined with a suitable speedup procedure leading to a drastic complexity reduction. The performance, measured within applications to image coding, prove the algorithm to be up to 50 times faster than the GLA method increasing the peak signal to noise ratio (PSNR) at knee of nearly 1 dB.

This paper addresses the framework of object-oriented image coding, describing a new algorithm, based on monodimensional Legendre polynomials, for texture approximation. Through the use of 1D orthogonal basis functions, the computational complexity which usually makes prohibitive most 2D region-oriented approaches is significantly reduced, while only a slight increment of distortion is introduced. With the aim of preserving the bidimensional intersample correlation of texture information as much as possible, suitable pseudo-bidimensional basis functions have been used, yielding significant improvements with respect to the straightforward 1D approach. The algorithm has been experimented with, for coding still images as well as motion compensated sequences, showing interesting possibilities of application for very low bit rate video coding.>

Very high compression in videophone coding can be reached successfully only if model-based segmentation is performed to allow suitable bit allocation. The exploitation of a priori knowledge suggests the application of fast and simple segmentation algorithms oriented at partitioning the image into variable resolution domains for subsequent texture encoding. In this paper we describe, together with the achieved preliminary results, a model-based approach to image segmentation relying on the estimation of the face symmetry axis and of the primary facial features. Through these parameters a flexible lattice is adapted frame by frame on the image, identifying a time-varying net of triangular patches whose texture is eventually encoded via Legendre basis functions. Target applications for videophone coding of QCIF color sequences at very low bitrate, less than 16 Kbit/sec., are foreseen.>

Signal sample quantization represents the basic operation of any system for digital signal processing and can be mathematically formalized as a least-distance application from the domain of input samples to a finite and fixed set of reproduction values generally called quantization levels, in case of scalar quantization, or reconstruction codewords in case of vector quantization. As the size of the reproduction set increases, the computational overhead introduced by the least-distance search procedure leads to a drastic reduction of the system performances. A fast and efficient implementation of the search algorithm represents a problem of key relevance, especially as far as high dimensional vector quantization is concerned. The algorithm presented in the paper employs a binary tree structure to address the reconstruction set according to the least-distance rule, providing a logarithmic reduction of the search complexity.>

In this paper a code-bit allocation policy is presented for the vector quantization of video data, based on the Human Visual System (HVS) frequency response. Discret Cosine Transform is employed for data decorrelation and a frame adaptive vector quantization mechanism is subsequently applied for the DCT coefficients encoding. HVS response is taken into account for the code-bit allocation by sizing the reconstruction look-up table of each vector quantizer. Before vector quantization, DCT coefficients are zig-zag reordered and thresholded; the quantized coefficients are subsequently run-length encoded. The resulting coder presents the basic characteristics of being block-oriented, layered and variable bitrate (VBR): promising results in applications within packet switched networks are forseen. Preliminary results are presented.