We propose iMoT, an innovative Transformer-based inertial odometry method that retrieves cross-modal information from motion and rotation modalities for accurate positional estimation. Unlike prior work, during the encoding of the motion context, we introduce Progressive Series Decoupler at the beginning of each encoder layer to stand out critical motion events inherent in acceleration and angular velocity signals. To better aggregate cross-modal interactions, we present Adaptive Positional Encoding, which dynamically modifies positional embeddings for temporal discrepancies between different modalities. During decoding, we introduce a small set of learnable query motion particles as priors to model motion uncertainties within velocity segments. Each query motion particle is intended to draw cross-modal features dedicated to a specific motion mode, all taken together allowing the model to refine its understanding of motion dynamics effectively. Lastly, we design a dynamic scoring mechanism to stabilize iMoT's optimization by considering all aligned motion particles at the final decoding step, ensuring robust and accurate velocity segment estimation. Extensive evaluations on various inertial datasets demonstrate that iMoT significantly outperforms state-of-the-art methods in delivering superior robustness and accuracy in trajectory reconstruction.

Despite remarkable progress in knowledge transfer across visual and textual domains, extending these achievements to indoor localization, particularly for learning transferable representations among Received Signal Strength (RSS) fingerprint datasets, remains a challenge. This is due to inherent discrepancies among these RSS datasets, largely including variations in building structure, the input number and disposition of WiFi anchors11An anchor describes a radio-emitting source. (e.g., WiFi access points, Bluetooth beacons.). Accordingly, specialized networks, which were deprived of the ability to discern transferable representations, readily incorporate environment-sensitive clues into the learning process, hence limiting their potential when applied to specific RSS datasets. In this work, we propose a plug-and-play (PnP) framework of knowledge transfer, facilitating the exploitation of transferable representations for specialized networks directly on target RSS datasets through two main phases. Initially, we design an Expert Training phase, which features multiple surrogate generative teachers, all serving as a global adapter that homogenizes the input disparities among independent source RSS datasets while preserving their unique characteristics. In a subsequent Expert Distilling phase, we continue introducing a triplet of underlying constraints that requires minimizing the differences in essential knowledge between the specialized network and surrogate teachers through refining its representation learning on the target dataset. This process implicitly fosters a representational alignment in such a way that is less sensitive to specific environmental dynamics. Extensive experiments conducted on three benchmark WiFi RSS fingerprint datasets underscore the effectiveness of the framework that significantly exerts the full potential of specialized networks in localization22Our code is available at https://github.com/Minh-Son-Nguyen/RSS_TL..

In this paper, we present all-embracing Transformers (AaTs) that are capable of deftly manipulating attention mechanism for Received Signal Strength (RSS) fingerprints in order to invigorate localizing performance. Since most machine learning models applied to the RSS modality do not possess any attention mechanism, they can merely capture superficial representations. Moreover, compared to textual and visual modalities, the RSS modality is inherently notorious for its sensitivity to environmental dynamics. Such adversities inhibit their access to subtle but distinct representations that characterize the corresponding location, ultimately resulting in significant degradation in the testing phase. In contrast, a major appeal of AaTs is the ability to focus exclusively on relevant anchors in RSS sequences, allowing full rein to the exploitation of subtle and distinct representations for specific locations. This also facilitates disregarding redundant clues formed by noisy ambient conditions, thus enhancing accuracy in localization. Apart from that, explicitly resolving the representation collapse (i.e., none-informative or homogeneous features, and gradient vanishing) can further invigorate the self-attention process in transformer blocks, by which subtle but distinct representations to specific locations are radically captured with ease. For that purpose, we first enhance our proposed model with two sub-constraints, namely covariance and variance losses at the Anchor2Vec. The proposed constraints are automatically mediated with the primary task towards a novel multi-task learning manner. In an advanced manner, we present further the ultimate in design with a few simple tweaks carefully crafted for transformer encoder blocks. This effort aims to promote representation augmentation via stabilizing the inflow of gradients to these blocks. Thus, the problems of representation collapse in regular Transformers can be tackled. To evaluate our AaTs, we compare the models with the state-of-the-art (SoTA) methods on three benchmark indoor localization datasets. The experimental results confirm our hypothesis and show that our proposed models could deliver much higher and more stable accuracy.

In this paper, we propose Anchor-agnostic Transformers (AaTs) that can exploit the attention mechanism for Received Signal Strength (RSS) based fingerprinting localization. In real-world applications, the RSS modality is inherently well-known for its extreme sensitivity to dynamic environments. Since most machine learning algorithms applied to the RSS modality do not possess any attention mechanism, they can only capture superficial representations, yet subtle but distinct ones characterizing specific locations, thereby leading to significant degradation in the testing phase. In contrast, AaTs are enabled to focus exclusively on relevant anchors at every Received Signal Strength (RSS) sequence for these subtle but distinct representations. This also facilitates the model to neglect redundant clues formed by noisy ambient conditions, thus achieving better accuracy in fingerprinting localization. Moreover, explicitly resolving collapse problems at the feature level (i.e., none-informative or homogeneous features) can further invigorate the self-attention process, by which subtle but distinct representations to specific locations are radically captured with ease. To this end, we enhance our proposed model with two sub-constraints, namely covariance and variance losses that are mediated with the main task within the representation learning stage towards a novel multi-task learning manner. To evaluate our AaTs, we compare the models with the state-of-the-art (SoTA) methods on three benchmark indoor localization datasets. The experimental results confirm our hypothesis and show that our proposed models could provide much higher accuracy.

With the exponential growth of facial authentications, the face anti-spoofing area has come to play an indispensable role as a shield, protecting those systems against facial impostures. However, because most current anti-spoofing technologies work with type-specific supervision, they are only effective in their respective spoof types, which means they are unlikely to prove robust for unidentified attack forms that are beyond their predefined supervised limitations. With this point in mind, we herein propose a novel Adversarial Distribution Generative Network (ADGN) that extends its spatial attention to a comprehensive global context, thus extensively raising the level of generality for unknown cases that inherently provide few or even no clues with which to learn. In this paper, we are more in favor of speculating on 3D mask attacks, where a great scarcity of prior knowledge is virtually inevitable due to their prohibitive costs. We also demonstrate the resilience of our proposed model and test it against publicly available datasets on both seen and unseen spoof scenarios. This intends to show how our model provides competitive detecting performance against a wide range of spoof types in comparison with previous state-of-the-art methods.

A premature crack is a significant indicator for early failure detection for underground polyvinyl chloride (PVC) water pipes. Using an array of strain gauges mounted on a pipe surface to monitor the strain of adjacent areas of a premature crack is a novel technology that has not been explored. To reduce the risks and up-front investments, we need a testbed to investigate, verify, and validate the innovative technology. However, establishing such a pipe monitoring testbed that covers completely realistic underground situations is challenging. The main reason lies in three main challenges: (i) mimicking the natural changes of water flowing in pipes; (ii) identifying the proper placement of strain gauges and detectable crack sizes; (iii) simulating the crucial underground conditions such as temperature and external stress. To this end, in this paper, we present a testbed to get more insights into the effects of different crack types on the pressure-strain characteristic in realistic conditions. In particular, we use pressure meters and water pumps to control the water flow (for challenge (i)); deploy various strain gauges types and induce cracks with different sizes (for challenge (ii)); fill the pipe with water at various temperatures and underground-like external stress (for challenge (iii)); Analyzing experimental results reveals useful hints for designing a realistic testbed, including but not limited to, the required distance among strain gauges, the influence of temperature and pipe axial stress. The dataset and analytic results of this work would provide more insights into how to design a realistic testbed for underground PVC water pipe crack detection.

The drinking water mains, mostly buried underground and stay there for decades, require proper maintenance to prevent failures. Among different kinds of material, polyvinyl chloride (PVC) has been widely used due to positive features, such as high durability, corrosion resistance, low price, and easy installation. To the best of our knowledge, this is the first article that reviews the inspection methods toward continuously monitoring the structural health of PVC drinking water mains. To understand which properties need inspecting, we first investigated the attributes that influence PVC pipe and joint failures. Then, we reviewed the methods that have already been applied or can inspect these influencing attributes. We categorized the prospects into five groups: 1) sound wave; 2) fiber optic sensing; 3) hydraulic monitoring; 4) multiple discrete sensors; and 5) other inline methods. Finally, we discussed the possibility and challenges in implementing these methods into a continuous monitoring system of PVC water mains for early failure warning. The result, which includes active sound wave, fiber optic sensing, hydraulic vibration, and multiple discrete sensors methods, can help future researchers select the appropriate methods to develop the continuous monitoring system for the PVC water mains.

WiFi fingerprinting using Convolutional Neural Networks (CNN) is one of the most promising techniques for indoor localisation due to the extraordinary performance of CNN in image classification. However, the performance of CNN is architecture dependant, and thus an architecture that works well in one case may not work in another, especially for the WiFi-based localisation problems. Most of the solutions use an existing hand-crafted architecture or a semi-automated CNN design for fingerprinting, which requires significant CNN expertise and time. Therefore, a satisfactory solution may not be guaranteed as it is challenging to design numerous possible architectures. In this work, we address this challenge by developing a framework that completely automates the CNN architecture design process. Our automated architectures based on VGG blocks have shown superior performance compared to standard architectures such as VGG-16. We further explore three different heuristics for automation: Bayesian optimisation, Hyperband, and Random Search and demonstrate their importance towards the automated CNN architecture development for WiFi fingerprinting. Experiments are conducted on real-world datasets and, a comparative study between our automated architecture and other models is presented. This work would, therefore, facilitate the CNN design for indoor localisation.

One of the most practical localization techniques is WLAN-based fingerprinting for location-based services because of the availability of WLAN Access Points (APs). This technique measures the Received Signal Strength (RSS) from APs at each indicated location to construct fingerprints. However, the collection of fingerprints is notoriously laborious and needs to be repeatedly updated due to the changes of environments. To reduce the workload of fingerprinting, we apply Deep Belief Networks to unlabeled RSS measurements to extract hidden features of the fingerprints, and thereby minimize the collection of fingerprints. These features are used as inputs for conventional regression techniques such as Support Vector Machine and K-Nearest Neighbors. The experiment results show that our feature representations learned from unlabeled fingerprints provide better performance for indoor localization than baseline approaches with a small fraction of labeled fingerprints traditionally used. In the experiment, our approach already improves the localization accuracy by 1.9 m when using only 10% of labeled fingerprints, compared to the closest baseline approach which used 100% of labeled fingerprints.

Bluetooth Low Energy (BLE) devices such as iBeacons have been popularly deployed for Location Based Services (LBS), including indoor infrastructure monitoring, positioning, and navigation. In these applications, the positions of iBeacons are assumed to be known. However, the location information is often unavailable or inaccurate as most iBeacons were deployed by different external parties. In addition, manual localizing the already-deployed iBeacons is costly and even impractical, especially in large-scale and complex indoor environments. Therefore, we propose a novel method, namely SomeBe, which can localize deployed iBeacons with a minimal effort and invasiveness to existing infrastructures. Specifically, our approach uses cooperative multilateration based on Received Signal Strength (RSS) of available smartphones and WiFi access points (APs) in the environment. Both Bluetooth signal strengths (between smartphones and iBeacons) and WiFi signal strengths (between smartphones and APs) are jointly employed in a single optimization cost function to surpass the local minima. Requiring that the positions of the APs are known only, the proposed cost function can also localize the iBeacons without knowing the positions of smartphones. To improve the localization accuracy, we employ a clustering method based on the RSS values for the coarse estimation of iBeacons' positions. SomBe also can be used to simplify iBeacon deployment as it can localize the iBeacons with a minimal effort. The performance evaluation results of our testbed experiments as well as realistic simulations show that SomBe outperforms non-cooperative approaches with 85% better in terms of accuracy.

In this article, we propose a novel cooperative hierarchical caching framework in a Cloud Radio Access Network (C-RAN), in which a new cloud-cache at Cloud Processing Unit (CPU) is envisioned to bridge the storage-capacity/delay-performance gap between the traditional edge-based and core-based caching paradigms. A delay-cost model is introduced and the cache placement problem is formulated that aims at minimizing the average delay-cost of content delivery in the network. Given the NP-completeness of the cache placement problem, we propose a low-complexity heuristic cache-management strategy comprising of a proactive cache-distribution algorithm and a reactive cache-replacement algorithm. Furthermore, a Cache-Aware Request Scheduling (CARS) algorithm is devised in order to optimize online the tradeoff between content download rate and content access delay. Via extensive numerical simulations-carried out using both real-world YouTube video requests and synthetic content requests-it is demonstrated that the proposed cache-management strategy outperforms traditional caching strategies in terms of cache hit ratio, average content access delay, and backhaul traffic load. Additionally, it is shown that the proposed CARS algorithm achieves superior tradeoff performance over traditional approaches that optimize either users' rate or access delay alone.

In this paper, we present an effective and low- cost wireless communication system for extremely long and narrow pipes that can replay the extant wire system in underground sensor network applications such as soil sampling and testing with the Cone Penetration Test (CPT), the most widely used underground sensor device. Different from existing in-pipe wireless techniques, we consider real-world pipelines that are very narrow and long. In particular, in our design data are first modulated at a commercial frequency and then converted to high frequency, between 14-15 GHz, to be transmitted along of the pipelines under the circular waveguide mode TM01. Especially, we design a cone-shaped antenna to overcome the aligning problem of feeds between the transmitter and receiver. To evaluate the applicability and efficiency of our design, we conduct realistic simulations as well as experiments with real prototypes. The results of experiments are consistent with our theoretical design and simulations and show that our proposed wireless system can transfer sensory data up to 20 m in narrow CPT pipes with a diameter of 17 mm when using the LoRa modulation with a transmitting power of 1 W, whereas existing underground radio techniques can transfer data from a depth of 2 m at maximum in the same condition. In our approach, it is also possible to add repeaters to extend the communication range when needed.

Self-organization is critical to enable novel indoor Location-Based Services (LBSs) for users and businesses in large, complex and unstructured buildings. Inspired by high densities of smartphones in public indoor spaces, in this paper we propose a self-organizing indoor localization approach that allows the use of available WiFi Access Points (APs) and iBeacons in the area to improve location accuracy and environment adaptability. Our approach is based on a semi-anchored localization that estimates the unknown location of smartphones, given known-location anchors (APs) and unknown-location anchors (iBeacons). We exploit the capabilities of Levenberg-Marquardt optimization algorithm to accurately estimate smartphone locations in realtime, in contrast to fingerprinting methods that require a tedious off-line training phase. Moreover, we use a clustering method based on the Received Signal Strength (RSS) values to obtain the initial estimated location for the optimization. We evaluate our approach using available APs and non-coordinated iBeacons in a large building to localize smartphones. The experimental results confirm that our self-organizing approach not only effortlessly estimates the position of mobile devices, but also provides a higher localization accuracy than other widely used approaches such as extant fingerprinting techniques for both scenarios, with and without iBeacons.

Existing approaches to activity recognition in smart homes mostly rely on supervised learning from well-annotated sensor data, acquired in a controled lab environment. However obtaining such labeled data in real home scenarios could be prohibitive due to either the privacy concerns of using cameras, or the low adherence of self reports done by home residents. Unsupervised learning, on the other hand, aims at discovering activities through applying fixed complexity models, yet assuming apriori knowledge of the number of activities. Again this is also a non-practical approach because the number of activities could vary drastically, even within a home. In this paper, we propose a novel practical unsupervised Bayesian nonparametric model to discover activities in smart homes, without prior assumption on the number of activities. Instead, our model can automatically infer such number only from sensor readings, thus it can be easily applied to any new home. We test our method on a public dataset and a dataset collected in our project. On the CASAS dataset, which has activity labels, our approach can achieve the performance close to the best of GMM and outperforms K-means. On our smart home dataset, the discovered activities are highly correlated with the typical daily routine of the resident. Such experimental results demonstrate the efficiency of our method for activity discovery in smart homes.

In the context of acoustic monitoring, the location of a sound source can be passively estimated by exploiting time-of-arrival and time-difference-of-arrival measurements. To evaluate the fundamental hardness of a location estimator, the Cramer-Rao bound (CRB) has been used by many researchers. The CRB is computed by inverting the Fisher Information Matrix (FIM), which measures the amount of information carried by given distance measurements. The measurements are commonly expressed as actual distances plus white noise. However, the measurements do include extra noise types caused by time synchronization, acoustic sensing latency, and signal-to-noise ratio. Such noise can significantly affect the performance and depend highly on the sensing platforms such as Android smartphones. In this paper, we first remodel the acoustic-based distance measurements considering such additive errors. Then, we derive a new FIM with the new statistical ranging error models. As a result, we obtain new CRBs for both non-cooperative and cooperative localization schemes that provide better insight into the causality of the uncertainties. Theoretical analysis also proves that the proposed CRBs for localization become the old CRBs when the additional errors are ignored, which gives a robust check for the new CRBs. Thus, the new CRBs can serve as a benchmark for localization estimators with both new and old measurement models. The new CRBs also indicate that there is room to improve current localization schemes, however, it is a daunting challenge.

Localization, a process of determining the position of a blind node, can be used in various applications. Signal-strength localization provides a low-cost and low-power solution to positioning. Signal-strength positioning approaches using fingerprinting or calibrated approaches require a time-consuming calibration phase. Existing self-calibrating approaches, which do not require a priori calibration, use a least-squares fitting model to determine both the position of the blind node as well as the optimal environmental parameters. In this paper, we propose an approach using the Product-Moment correlation between the measured signal strength and the estimated signal strengths. Such approach does not require estimation of the environmental parameters or prior calibration and outperforms existing self-calibrating least-squares approaches. We compare our approach to existing least-squares calibration-free positioning approaches. Moreover, we look at the Cramer-Rao Bound (CRB) of signal-strength localization and using simulations we show that the product-moment correlation outperforms least-squares approaches and follows the CRB closely. Simulation and evaluation using a real-world experiment dataset show the product-moment approach significantly outperforms least-squares approaches. The product-moment approach follows the CRB much more closely and achieves up to twice more accurate positions in certain scenarios. When the error ratio increases and the number of reference positions stays fixed at 6, the product-moment approach scores 20% more accurate positions. In the cooperative localization scenario, the product-moment correlation performs 40% better.

The emerging wave of technology in human-centric devices such as smart phones, tablets, and other small wearable sensor modules facilitates pervasive systems and applications to be economically deployed on a large scale with human participation. To exploit such environment, data gathering and dissemination based on opportunistic contact times among humans is a fundamental requirement. To tackle the lack of contemporaneous end-to-end connectivity in Delay-tolerant Networks (DTNs), most current algorithms assess the probability of the contact times to gradually convey a message towards its destination. These contact-based approaches do not perform well when historical locations of nodes have mixture distribution. In this paper, we formulate routing problems in spatial and spatiotemporal domains as an online unsupervised learning problem given location data. The key insight is that nodes frequently appearing nearer the message destinations are regarded as possessing higher delivery probability even if they have low contact times. We show how to solve the formulated problems with two basic algorithms, Location-Mean and Location-Cluster, by estimating the means of historical locations to calculate delivery probability of nodes. To our best knowledge, this is the first work to tackle DTN routing problem using online unsupervised learning on geographical locations. In the context of human mobility, simulation results of the Location-Mean algorithm show that the online unsupervised learning approach given node locations achieves better routing performances in term of delivery ratio, latency, transmission cost, and computation efficiency compared to the contact-based approach.