ABSTRACT. This paper analyses the track-level detection of ghost tracks in multitarget tracking with a known reflection surface. In a real-world target tracking problem, the number of targets in the surveillance is unknown to the platform. Thus, the tracker will be inadequate to distinguish the direct target return from the multipath return during the track initialization. Therefore, ghost tracks can be created with multipath measurements when they are considered direct path measurements. Even though the possible multipath measurement could be predicted for the existing tracks at a given instance, it is hard to decide whether the detected track is a multipath or a new target. Thus, a sequence of time instances needs to be considered to determine the track status. In this work, we propose a classification model to classify a track as either a multipath or direct path using an LSTM network with sequential data. Additionally, the performance of the proposed approach is compared with four other algorithms using a simulation-based dataset.

ABSTRACT. Maneuvering target tracking, where the system undergoes abrupt changes in the underlying motion model, can be challenging. We propose a model-based deep learning approach for prediction of maneuvering targets to exploit partial knowledge of the system physics-based models during training, without requiring an explicit characterization or fine tuning of model parameters. We formulate a supervised training scheme to learn the dynamics of state-space models and capture the jump processes governing model transitions by minimizing the prediction loss of an encoder-decoder network from model-based generated data. The effectiveness of the proposed method is demonstrated in two maneuvering target tracking scenarios using synthetic and real-world test data. The results show that the model-based encoder-decoder network achieves notably improved performance in terms of target prediction compared to conventional multiple-model solutions, especially when facing model inaccuracies, jumps, and dominant nonlinearities during target maneuvers.

ABSTRACT. Wi-Fi devices can effectively be used as passive radar systems that sense what happens in the surroundings and can even discern human activity. We propose, for the first time, a principled architecture which employs Variational Auto-Encoders for estimating a latent distribution responsible for generating the data, and Evidential Deep Learning for its ability to sense out-of-distribution activities. We verify that the fused data processed by different antennas of the same Wi-Fi receiver results in increased accuracy of human activity recognition compared with the most recent benchmarks, while still being informative when facing out-of-distribution samples and enabling semantic interpretation of latent variables in terms of physical phenomena. The results of this paper are a first contribution toward the ultimate goal of providing a flexible, semantic characterisation of black-swan events, i.e., events for which we have limited to no training data.

ABSTRACT. and civilian surveillance and protection systems. The complementarity and redundancy of sensors enables detecting, tracking and classifying Unmanned Aerial Vehicles (UAVs), further assessing their possible threat and planning the proper counter-measure. Despite efforts to develop dedicated systems (sensors, processing, hard and soft kill systems), counter Unmanned Aerial Systems (CUAS) are often deployed in complex areas, which requires an adequate understanding of the situation including the UAVs behaviour for an appropriate response. The evaluation of CUAS should thus consider not only its ability to provide a tactical picture of sufficient quality, but also to provide semantic information, establish possible links between objects and contextualize its behaviour. In this paper, we set up the basics for an evaluation platform covering the lower levels but also higher levels of the JDL (Joint Directory Laboratory) functional model of information fusion. We consider the six classical criteria of a tactical picture (used in level 1) and extend them to higher level tasks. Evaluation criteria are aligned with the URREF (Uncertainty Representation and Reasoning Evaluation Framework) ontology evaluation criteria. Furthermore, we propose some metrics to quantify such evaluation criteria. The framework is illustrated in a CUAS scenario, with data provided by the LEXULTOR platform. Several fusion solutions are compared which differ in the way uncertainty is captured and provided to the user

ABSTRACT. ChatGPT, OpenAI’s chatbot, has gained consider-able attention since its launch in November 2022, owing to its ability to formulate articulated responses to text queries and comments relating to seemingly any conceivable subject. As impressive as the majority of interactions with ChatGPT are, this large language model has a number of acknowledged shortcomings, which in several cases, may be directly related to how ChatGPT handles uncertainty. The objective of this paper is to pave the way to formal analysis of ChatGPT uncertainty handling. To this end, the ability of the Uncertainty Representation and Reasoning Framework (URREF) ontology is assessed, to support such analysis. Elements of structured experiments for reproducible results are identified. The dataset built varies Information Criteria of Correctness, Non-specificity, Self-confidence, Relevance and Inconsistency, and the Source Criteria of Reliability, Competency and Type. ChatGPT’s answers are analyzed along Information Criteria of Correctness, Non-specificity and Self-confidence. Both generic and singular information are sequentially provided. The outcome of this preliminary study is twofold: Firstly, we validate that the experimental setup is efficient in capturing aspects of ChatGPT uncertainty handling. Secondly, we identify possible modifications to the URREF ontology that will be discussed and eventually implemented in URREF ontology Version 4.0 under development.

ABSTRACT. Subjective Bayesian networks (SBN) extend Bayesian Networks (BN) by combining the factored joint probability representation of BNs with Subjective Logic, in such that second-order uncertainty (i.e., the uncertainty about a probability distribution -- as opposed to the uncertainty about the event itself) is represented as opinions. While SBNs provide a strong formalism for treating the uncertainty in a higher context, the literature lacks support for extensive software implementations focused on compatibility with present software solutions or standards, and applications in a systemic software/knowledge engineering process. Our work explores the structural congruence between BN and SBN (in terms of software data structure) and a semantic bijection between Subjective Logic opinions to Beta/Dirichlet distributions to introduce a SBN reasoning framework that targets on effectively reusing the existing BN solutions. Particularly, we developed two inference algorithms that apply a Monte-carlo method to existing BN inference algorithms (we chose the Junction Tree algorithm, for test), respectively for batch and interactive SBN reasoning. A method for translating an evidence in SBN to uncertain (virtual) evidence in BN is also presented. We also developed a Java component to demonstrate the reusability as a software asset, and developed a case study of Mission Impact Assessment of Unmanned Aerial Vehicles transporting critical items between hospitals, in order to illustrate its applicability in a systemic knowledge engineering process.

ABSTRACT. Safe and reliable state estimation techniques are a critical component of next-generation robotic systems. Agents in such systems must be able to reason about the intentions and trajectories of other agents for safe and efficient motion planning. However, classical state estimation techniques such as Gaussian filters often lack the expressive power to represent complex underlying distributions, especially if the system dynamics are highly nonlinear or if the interaction outcomes are multi-modal. In this work, we use normalizing flows to learn an expressive representation of the belief over an agent's true state. Furthermore, we improve upon existing architectures for normalizing flows by using more expressive deep neural network architectures to parameterize the flow. We evaluate our method on two robotic state estimation tasks and show that our approach outperforms both classical and modern deep learning-based state estimation baselines.

ABSTRACT. This paper presents recursive two-filter smoothing (TFS) in the criterion of maximizing the correntropy (MC) instead of minimizing the mean square error, to pursue robustness for outlier rejections caused by non-Gaussian noises and obtain high-precision state estimate, which is motivated by non-cooperative target backtracking. Here, non-cooperative target tracking often needs to consider non-Gaussian noises. The MC-based recursive TFS (abbreviated as MRTFS) is put forward, where both the forward and backward filters are performed independently and recursively in the criterion of MC. Meanwhile, an MC-based fusion rule is further designed to obtain the final smoothed estimate by fusing the forward filtered estimate and backward predicted estimate step by step, in order to improve estimation accuracy. A target backtracking example with non-Gaussian noises is simulated to show the advantage of estimation accuracy of the proposed MRTFS over Kalman filter/smoothers, MC-based Kalman filter/Rauch-Tung-Striebel smoother, in terms of different kernel bandwidths and levels of process noises.

ABSTRACT. Extended Object Tracking (EOT) is a useful technique for achieving situational awareness in autonomous vehicle systems. The EOT problem is to both estimate the kinematics and dimensions of an object based on high-resolution measurements. In the case of laser measurements or other types of measurements that correspond to points on the object's boundary, the true measurement model of the EOT problem is based on an implicit equation for the measurement coordinates. This intrinsic implicity is often not addressed directly in several EOT models found in the literature. In this paper, the EOT problem is reformulated as a least square minimization problem without compromising the original implicit measurement model by introducing an extra variable for each measurement. In addition, this new least squares formulation allows considering measurements and state variables for a whole time window, and not just a single time step. An EOT algorithm based on solving the derived least squares minimization problem is proposed and tested with simulated scenarios.

ABSTRACT. In this research, we aim to develop a new, low- complexity, robust parameter estimation method for periodic parameter estimation. In many practical parameter estimation problems, such as phase, frequency, and direction-of-arrival (DOA) estimation, the observation model is periodic with respect to the unknown parameters, and thus, the appropriate estimation criterion is periodic in the parameter space. However, iterative estimation methods, such as the Fisher-Scoring method, do not take into consideration the periodic information in order to improve the accuracy of the estimation. In this paper, we present a new iterative method - periodic Fisher-Scoring, which takes into account the signal’s periodic properties through the utilization of the cyclic Cram´er-Rao bound (CRB). The cyclic CRB is a lower bound on the mean cyclic error (MCE) of unbiased estimators and, thus, is more appropriate for the derivation of the iterative method. In addition, the periodic Fisher-Scoring uses the modulo 2π operator at each iteration. The performance of the proposed Fisher-Scoring estimator is compared to the performance of the conventional Fisher-Scoring method. We also compare the results with the CRB and the cyclic CRB. Simulation results for DOA estimation in seismic arrays show that the proposed periodic Fisher-Scoring estimator has a lower MCE compared to the conventional Fisher-Scoring estimator. The performance improvement is more significant around the edges of the range [−π, π] and under the misspecified model, i.e. under the mismatched assumption of white noise.

ABSTRACT. Human-wildlife conflicts are a global problem which is central to the Global Goal 15 (life on land). One particular case is elephants, that can cause harm to both people, property and crops. An early warning system that can detect and warn people in time would allow effective mitigation measures. The proposed method is based on a small local network of geophones that sense the seismic waves of elephant footsteps. It is known that elephant footsteps induce low frequency ground waves that can be picked up by geophones in the ground. First, a method is described that detect the particular signature of such footsteps, and then the detections are used to estimate the direction of arrival (DOA). Finally, a Kalman filter is applied to the measurements in order to track the elephant. Field tests performed at a local zoo shows promising results with accurate DOA estimates at 15 meters distance and acceptable accuracy at 40 meters.

ABSTRACT. We consider the problem of collaborative bearing estimation using a method with historic roots in set theoretic estimation techniques. We refer to this method as the Convex Combination Ellipsoid (CCE) method and show that it provides a less conservative covariance estimate than the well known Covariance Intersection (CI) method. The CCE method does not introduce additional uncertainty that was not already present in the prior estimates. Using our proposed approach for collaborative bearing estimation, the nonlinearity of the bearing measurement is captured as an uncertainty ellipsoid thereby avoiding the need for linearization or approximation via sampling procedures. Simulations are undertaken to evaluate the relative performance of the collaborative bearing estimation solution using the proposed (CCE) and typical (CI) methods.

ABSTRACT. Event detection is a crucial and challenging subtask of event extraction, which suffers from the ambiguity of natural language and in real-world scenarios. Existing work mainly focuses on using textual context information, while event information may co-exist in rich multimedia content (e.g., news articles usually come with textual documents and images/videos). Multi-modal event detection models require a large amount of labeled data. This hinders their real-world practicality where domain-specific multi-modal data are scarce. In this paper, we present a novel domain-adaptive visually fused event detection approach that can be trained on relatively few labeled paired image-text data points. Specifically, we introduce a visual imagination method, that can be customised on a specific domain and synthesize images from text in the absence of visual images. In doing so, our model can leverage the capabilities of pre-trained vision-language models and can be trained on few-shot data. This also allows for effective inference where only single-modality data (i.e., text) is available. The experimental evaluation on the benchmark M2E2 dataset shows that our model outperforms existing state-of-the-art models.

ABSTRACT. Multi-frame detection has been widely researched in the scenario where the target signal-to-noise is low.However, it becomes a challenging problem under heterogeneous clutter environment. As strong clutter energy is accumulated along with the target in multiple frames, low SNR targets are still annihilated in clutter. To achieve effective clutter suppression and dim targets detection, a novel multi-frame procedure for energy accumulation under heterogeneous clutter is proposed in this paper. The presented architecture concerns a Space-Time Adaptive Processing (STAP) processor and a multi-frame detector. The STAP processor calculates the clutter covariance matrix using multi-frame training samples near the cell under test and extracts data contaminated by the target component. The multi-frame detector is developed to detect dim targets and output estimated target track sequences. Finally, simulation results are given to demonstrate the efficacy of the proposed algorithm.

ABSTRACT. The problem of biased measurements in bearings-only target motion analysis (BOTMA) is the topic of this article. More precisely, we thoroughly establish the observability condition of the trajectory of the target together with the bias (assumed constant during the whole scenarios).

ABSTRACT. Meta-learning, decision fusion, hybrid models, and representation learning are topics of investigation with significant traction in time-series forecasting research. Of these two specific areas have shown state-of-the-art results in forecasting: hybrid meta-learning models such as Exponential Smoothing - Recurrent Neural Network (ES-RNN) and Neural Basis Expansion Analysis (N-BEATS) and feature-based stacking ensembles such as Feature-based FORecast Model Averaging (FFORMA). However, a unified taxonomy for model fusion and an empirical comparison of these hybrid and feature-based stacking ensemble approaches is still missing. This study presents a unified taxonomy encompassing these topic areas. Furthermore, the study empirically evaluates several model fusion approaches and a novel combination of hybrid and feature stacking algorithms called Deep-learning FORecast Model Averaging (DeFORMA). The taxonomy contextualises the considered methods. Furthermore, the empirical analysis of the results shows that the proposed model, DeFORMA, can achieve state-of-the-art results in the M4 data set. DeFORMA, increases the mean Overall Weighted Average (OWA) in the daily, weekly and yearly subsets with competitive results in the hourly, monthly and quarterly subsets. The taxonomy and empirical results lead us to argue that significant progress is still to be made by continuing to explore the intersection of these research areas.

ABSTRACT. In aviation there are many safety-critical domains where reliable safety systems are essential to prevent any kind of hazard. This paper focuses on airport aprons, where currently used holding point protection systems have shown to be not faultless, sometimes leading to avoidable accidents. One way to avoid such accidents is by means of innovative sensor technology, in our case, magnetometers, i.e. sensors measuring the distortion of the earth's magnetic field by metallic objects. The main goal is to use the magnetometry data to detect passing aircraft and to capture their geometrical pattern as well as to estimate their motion vector. Therefore, we present a spatio-temporal cluster fusion and an event fusion algorithm. The cluster fusion can be applied as a post-processing step to any spatio-temporal clustering method and is able to more accurately represent aircraft patterns by integrating expert knowledge into the fusion process. In this context, we present a spatio-temporal cluster tree representation for a fast and accurate estimation of the motion vector. Finally, the data-driven event fusion is able to separate detected aircraft crossings into separate events by employing domain-knowledge. In future work, we aim to come up with a framework making use of the cluster results and estimated motion vector to classify and infer the position of an aircraft, before this is deployed as a real-time application.

ABSTRACT. Research in Natural Language Processing, biomedicine, and computer vision achieved excellent results in machine learning due to the success of the Transformer-based models. However, these excellent results depend on the labeled high-quality and large-scale datasets. If one of these requirements is not met, the model may lack generalization ability, and its performance will be unsatisfactory. To address these issues, this research proposes a Generic Multimodal Gradient-Based Meta Framework (GeMGF) trained from scratch to avoid language bias, learns from a few data, and reduces the model degradation trained on a finite dataset. GeMGF was evaluated using the benchmark dataset CUB-200-2011 for the text and image classification tasks. The results show that GeMGF outperforms the state-of-the-art models with 93.2% accuracy. GeMGF is simple, efficient, and adaptable to other data modalities and fields.

ABSTRACT. AI-based decision support solutions require life cycles that adequately address critical steps, such as (i) finding suitable AI methods for the problem at hand, (ii) preparing and executing adequate data acquisition processes and (iii) tractable evaluation of the overall solution. Understanding the data generating processes is key in achieving this. Training and test data can be seen as a result of a causal data generation process, a sampling process in which the data is collected from different sources that are influenced by multiple interdependent phenomena. This is represented by a Qualitative Model of Data Generation Processes (QM-DGP), a causal graphical model. QM-DGP facilitates analysis of the complexity of the underlying data generating processes that can inform the development of trustable ML-based solutions in multiple ways. Firstly, this analysis is the basis for the determination of the required complexity of the ML models. Secondly, it facilitates the determination of the quantities of training data supporting good learning results. Thirdly, it can provide guidance for a systematic simplification of the models, supporting tractable solutions without significantly reduced performance. The construction of QM-DGP and the analysis benefit from sound theoretical concepts, such as d-separation and I-Maps. The approach is agnostic to the types of models and machine learning. Preliminary experimental results with simulated data indicate that the approach can be effective in predicting the required quantities of training data and the determination of the model complexity using different types of models.

ABSTRACT. Classification of vessels is a major task in many duties of maritime surveillance. In previous work, ad hoc approaches for classifying maritime vessels were applied to large sets of single AIS messages. The resulting performance has been explored based on real-life datasets in different maritime areas. The approaches were used in a particular application scenario of detecting AIS spoofing in fishery scenarios. Seeing good classification results, in particular by Random Forests, the questions arise whether these at least partly resulted from overfitting phenomena and which of the feature of classification in this problem are most relevant. To address these questions in the current contribution, AIS messages are aggregated per vessel and by the hour by statistical means providing a new AIS feature representation as a basis for classification. By providing a new data representation to the Random Forest classifier, we evaluate classification performance. Then we restrict the presented data to the semantic feature subgroups of the vessel's positional, course, speed, and dimensions data individually and identify the key features of the classification. The results show that in our exemplary maritime experiments, classification relies mostly on vessels' positional and dimensional information.

ABSTRACT. The quotient of two multivariate Gaussian densities can be written as an unnormalized Gaussian density, which has been applied in some recently developed multiple-model fixed-interval smoothing algorithms. However, this expression is invalid if instead of being positive definite, the covariance of the unnormalized Gaussian density is indefinite (i.e., it has both positive and negative eigenvalues) or undefined (i.e., computing it requires inverting a singular matrix). This paper considers approximating the quotient of two Gaussian densities in this case using two different approaches to mitigate the caused numerical problems. The first approach directly replaces the indefinite covariance of the unnormalized Gaussian density with a positive definite matrix nearest to it. The second approach computes the approximation through solving, using the natural gradient, an optimization problem with a Kullback-Leibler divergence-based cost function. This paper illustrates the application of the theoretical results by incorporating them into an existing smoothing method for jump Markov systems and utilizing the obtained smoothers to track a maneuvering target.

ABSTRACT. A new class of iterated linearization-based nonlinear filters, dubbed dynamically iterated filters, is presented. Contrary to regular iterated filters such as the iterated extended Kalman filter (IEKF), iterated unscented Kalman filter (IUKF) and iterated posterior linearization filter (IPLF), dynamically iterated filters also take nonlinearities in the transition model into account. The general filtering algorithm is shown to essentially be a (locally over one time step) iterated Rauch-Tung-Striebel smoother. Three distinct versions of the dynamically iterated filters are especially investigated: analogues to the IEKF, IUKF and IPLF. The developed algorithms are evaluated on 25 different noise configurations of a tracking problem with a nonlinear transition model and linear measurement model, a scenario where conventional iterated filters are not useful. Even in this "simple" scenario, the dynamically iterated filters are shown to have superior root mean-squared error performance as compared to their respective baselines, the EKF and UKF. Particularly, even though the EKF diverges in 22 out of 25 configurations, the dynamically iterated EKF remains stable in 20 out of 25 scenarios, only diverging under high noise.

ABSTRACT. Nonlinear measurement models pose a challenge to linear filters. The ensemble Kalman filter (EnKF) is a popular choice despite its tendency to diverge in systems with highly accurate, highly nonlinear measurements. In this work, we present the Bayesian Recursive Update EnKF (BRUEnKF): a novel EnKF that employs the Bayesian Recursive Update Filter (BRUF) measurement update. The BRUF divides the the extended Kalman filter (EKF) update into an integer number of steps, allowing for the recomputation of the measurement Jacobian at regular intervals. We adapt the BRUF update for an ensemble filter, taking advantage of the EnKF's numerical covariance computation at each update step. The BRUEnKF is shown to improve on the EnKF for systems with range measurements.

ABSTRACT. In directional statistics, the von Mises-Fisher (vMF) distribution has long been a mainstay for inference with data on the unit hypersphere. The performance of statistical inference based on the vMF distribution, however, may suffer when there are significant outliers and noise in the data. Based on an analogy of the median as a robust measure of central tendency and its relationship to the Laplace distribution, we propose the spherical Laplace (SL) distribution, a novel probability measure for modelling directional data. In this paper, we study foundational properties of the distribution such as theoretical results on maximum likelihood estimation and a sampling scheme for probabilistic inference. We derive efficient numerical routines for parameter estimation in the absence of closed-form formula. An application of model-based clustering is considered under the finite mixture model framework. Our numerical methods for parameter estimation and clustering are validated using simulated and real data experiments.

ABSTRACT. We present a novel target tracking algorithm designed to track a target in the presence of wake clutter. What distinguishes the method from previous ones is that it also models the existence of the target and exploits the information provided by the wake measurements for this purpose. We present two ways of modeling the wake and evaluate the algorithm performance on simulated data. Results show that the method improves upon comparable target tracking methods when wake measurements are present.

ABSTRACT. In around the corner radar, particle filter tracking can help to improve the estimation accuracy of a Non-Line of Sight (NLOS) target position. However, in case of using one single receiving antenna, the measurement model only contains multipath delay information. Due to the lack of information in the model, particles corresponding to ambiguous positions can be assigned high weight values. This can leads the particle cloud to diverge from the actual target trajectory and requires several time steps to converge on it again. In this work, our proposed solution consists in exploiting additional Direction-of- Arrival (DoA) information provided by an array of receiving antennas in the particle filter framework. By simulation and experimentation, we will show that the proposed solution allows to solve the aforementioned problem, and thus achieves better localization results compared to single antenna-based tracking as well as non-tracking localization algorithm.

ABSTRACT. Decentralized data fusion algorithms are fundamentally built on the exchange of estimates and covariance matrices between the individual components. This leads to a high volume of data, mainly caused by the covariance matrices, which can be problematic, especially in environments with limited bandwidth. In order to guarantee the proper functioning of decentralized estimation algorithms, data reduction methods for covariance matrices must ensure that the reduced matrices are conservative, i.e., do not underestimate the actual uncertainty. Motivated by these considerations, this paper presents an elementwise event-triggered method for the data-reduced transmission of covariance matrices that takes into account the aforementioned condition concerning uncertainty. For this purpose, several event triggers are proposed and, based on the event data and diagonal dominance, upper bounds for the actual covariance matrices are derived. An investigation of the data reduction and its influence on the estimation results is performed in a decentralized tracking scenario. The results show that substantial data reduction is possible with only minor losses in estimation quality.

ABSTRACT. In the context of single air picture fusion based on tracks produced by several heterogeneous C2 system, this paper presents our original algorithmic approach and results of the decentralized fusion process. In this applicative context, the decentralized fusion node receives potentially very numerous (thousands) tracks from all registered C2. Some registered C2 are inconstant in terms of performance and accuracy, resulting in heterogeneous data and possibly in ill associations or estimations; therefore tracks and covariances are not always reliable especially in regards to consistency and speed. A fast, parallelizable and robust track association algorithm is presented, which cope with these C2 tracks information. Track association is made by partitioning a track graph. A tactical situation maintenance strategy inspired by the state of the art in MHT is proposed.

ABSTRACT. In this work, we consider a binary hypothesis testing problem involving human decision-makers. Due to the nature of human behavior, human decision-makers observe the phenomenon of interest sequentially up to a random length of time. The humans use a belief model to accumulate the log-likelihood ratios until they cease observing the phenomenon. The belief model is used to characterize the perception of the human decision-maker towards observations at different instants of time, i.e., some decision-makers may assign greater importance to observations that were observed earlier, rather than later and vice-versa. We further consider the performance of a group of humans using a global decision-maker that fuses human decisions using the Chair-Varshney rule. When the number of observations that were used by the humans to arrive at their respective decisions are available to the FC, the weights in the Chair-Varshney rule are modified to include this information in the decision fusion rule. Numerical and Simulation results are presented to corroborate and validate theoretical results.

ABSTRACT. In this paper, we develop an approach for measurement-to-track (M2T) interframe association in the presence of camera panning and zooming from drone-captured video. Standard M2T methods assume that the target motion can be used to predict the "measurement association regions" for the bounding boxes. If there is a sudden state change due to camera shift (panning) and zooming, it will lead to incorrect associations and poor tracking results. To solve this, the zoom ratio and panning in 2D coordinates are used to describe the camera motion parameters in each frame. The estimated parameters are obtained by a grid search combined with global assignment or directly solved using the linear least squares method, which is also combined iteratively with assignment. The goal is to achieve correct M2T associations by adjusting the predicted measurements using the estimated camera parameters. These "improved" predictions can also be used to update the target state with filtering algorithms. Frames with panning or/and zooming from real data are used to illustrate the effectiveness of the proposed methods and compared with the validation gate method based on inflated covariances.

ABSTRACT. Electro-optical/Infrared (EO/IR) camera systems are commonly used in target detection and tracking applications. Such camera systems typically comprise a suite of sensors such as narrow/wide Field of View (FOV) cameras that provide target-originated angular measurements. To estimate the target position in Cartesian space, existing techniques in literature employ the non-linear measurement mapping from the Focal Plane Array (FPA) to azimuth and elevation space. A common assumption made in using this conversion is that azimuth and elevation measurement errors have the same standard deviation, are uncorrelated and are uniform across the camera’s FOV. This paper presents an approach to derive the azimuth and elevation statistics including the cross-correlation of their errors. This approach converts the raw target measurements and their covariance in the image space (FPA) to the angular space for subsequent use in Cartesian state filtering. This conversion has been validated to be unbiased and consistent, and results show that the Line of Sight (LOS) angle error variances and their correlations are in fact variable, with magnitudes dependent on the target’s location in the FPA. The correct LOS angle covariance matrices should be used in Cartesian state estimation and fusion rather than the assumed constant angle variances and uncorrelated errors between the azimuth and elevation.

ABSTRACT. Providing efficient yet accurate statistical models is a challenging problem in many applications. When elementary models are not sufficiently descriptive, mixtures of densities can be used. A complexity management issue arises when mixture models are employed: the number of components should be a trade-off between the complexity and the accuracy of the model. However, in general, it is not obvious how to determine the right number of mixture components for a specific application. In a previous work, theoretical foundations to address such a topic have been laid, grounded on the use of the Composite Transportation Dissimilarity between mixtures, and a preliminary criterion to manage the complexity of a mixture model has been proposed. In this paper, additional theoretical insights are provided that allow to formulate a novel adaptive mixture reduction algorithm. Numerical tests show that in most cases the new algorithm constitutes a significant improvement over the previous one.

ABSTRACT. Test and Evaluation for verification and validation (V&V) of sensor data fusion techniques utilize methods of uncertainty analysis. The Uncertainty Representation and Reasoning Evaluation Framework (URREF) ontology identifies many attributes of metrics (i.e., semantic meaning, object metrics, and subjective quality). With the growing interest in artificial intelligence (AI) due to large data corpus access, fast compute power, and machine/deep learning (ML/DL) techniques; V&V of these methods are needed. In this paper, the enhancement of the URREF to utilize a risk assessment for decision is demonstrated towards analysis/alignment of ML/DL methods that utilize multi-modal data fusion. Evidential reasoning is considered in the use case to provide data handing reliability source and processing credibility to measure decision risk in a maritime domain awareness scenario.

ABSTRACT. This paper introduces the concept of cross-entropy and relative entropy of two basic belief assignments. It is based on the new entropy measure presented recently. We prove that the cross-entropy satisfies a generalized Gibbs-alike inequality from which a generalized Kullback-Leibler divergence measure can be established in the framework of belief functions. We show on a simple illustrating example how these concepts can be used for decision-making under uncertainty.

ABSTRACT. Track initiation in multi-object tracking for groups of objects traveling close to each other may require considering many unlikely tracks in close proximity. Limiting track numbers by not initiating within the gate, as is commonly done, does not work well in these scenarios. With many tracks, computing the exact marginal track to measurement probabilities in a joint integrated probabilistic data association (JIPDA) is computationally expensive.

In addressing this problem, we consider the linear multitarget (LM) IPDA, which is a linear approximation of the data association in JIPDA. Here, we formulate it as a Poisson point process (PPP) approximation of the track measurement densities with a particular intensity function. Given our focus on track initialization, we devise how to use the LM approximation on only a subset of the tracks, while the other tracks can thereafter be treated as in JIPDA. This gives a novel new approximation of the track to measurement data association probabilities which we term LMS.

Simulations show that using the LM technique is superior to the PHD filter in terms of posterior track existence probability. Further, it is seen that the PHD as an intensity function in LM performs worse than the original LM in terms of the data association probabilities. The LMS data association probabilities are also shown to typically have better worst-case errors than the original LM and loopy belief propagation (LBP). In terms of the GOSPA metric, initiating tracks on every measurement using LMS gives more reliable and faster track initialization for objects appearing close to already established tracks.

ABSTRACT. The problem of association of measurements acquired by passive bearings-only sensors in two dimensional (2D) plane is addressed in this paper. This problem can be formulated mathematically as a multidimensional assignment (MDA) problem with two steps of cost calculation and optimization. Compared with the optimization step, the cost calculation consumes more time (at least 80% of the total time for solving the MDA problem). In order to reduce the computational requirements of the assignment costs of infeasible associations, a practical coarse gating strategy is proposed. First, two bearing measurements from different sensors are used to predict the bearing measurements of other sensors. Then, infeasible associations can be identified using gates centered on the predicted bearing measurements. This strategy is also extended to the measurement data association of heterogeneous sensors. Numerical results verify the effectiveness of the proposed strategy.

ABSTRACT. We examine prototypical versions of the probabilistic data association (PDA) filter and the joint PDA filter developed in the Soviet Union in 1972. These filters, developed by Y.S. Achkasov, are previously undiscussed in the conventional Englishlanguage multitarget tracking (MTT) literature, despite being translated into English from their original Russian soon after publication. We place this work in its appropriate historical context and demonstrate the equivalence of the proto-(J)PDA with the conventional forms of these algorithms, as well as discuss areas where this work anticipates multiple decades of MTT filter development.

ABSTRACT. Inertial sensors provide a low-power and high-fidelity pathway for state estimation and sensor fusion. Inertial measurement units now feature on-chip processors for ultra-low-power information fusion, signal processing, and neural network-based classification at the extreme edge. However, accounting for domain shifts, personalized inference requirements, and application diversity makes adopting existing learning-enabled on-device training, classification, and fusion frameworks for on-sensor processors difficult. This paper introduces a method for personalized and on-device learning for on-chip classification, inference, and information fusion applications. The proposed framework automatically segments and stores quantized gravity vector image templates and axes variance information of motion artifacts during training. During inference, templates created from the time-series windows are matched against uniform blurred templates using the universal image quality index. An adaptive rep counting module robust to varying motion primitives counts repetitions of matched motion primitives. The framework requires no human-engineered parameters and allows for the personalization and addition of new motion artifacts. Our framework recognizes human activities with 96.7% test accuracy and achieves an average rep count error of 0.44, while reducing the memory usage by 1000-2000x over existing tiny machine learning on-device learning techniques, allowing on-sensor learning and inference under 8 KB of memory.

ABSTRACT. Power grid networks eventually require guyed towers as support structures for transmission lines. In these cases, the cables that support the structure can experience long-term degradation as a result of environmental forces. Long-term degradation can result in tower collapse leading to power outages in essential public services. We propose a Structure Health Monitor (SHM) system to improve transmission line reliability. It is based on data fusion using machine learning algorithms that monitor acceleration signals collected from multiple locations of the guyed tower that can allow the identification of loose cables and the amount of tightness. We select the most relevant individual measurable properties as a set of uncorrelated input sources, which can be in the time or frequency domain. The overall results for loose cable estimation of all investigated methods show a balanced accuracy in the range of 85% up to 96% and identification of tightness shows values between 91% and 96% for respectively inference using selected features set by investigated methods and all features available.

ABSTRACT. This paper presents an eXplainable Deep Reinforcement Learning (XDRL) based strategy for solving the proposed problem of fleet-level aircraft maintenance scheduling (AMS) optimization. The problem processes the information of the aircraft’s initial status, mission requirements, maintenance resource capacity, and other operational constraints to schedule sequential aircraft maintenance activities for a specified time horizon based on considering both mission readiness and cost reduction. An RL environment tailored for this problem, named AMS-Gym, was constructed using the OpenAI Gym toolkit. It is designed to be flexible to extend to more complicated scenarios with additional explanation capabilities. The explainable RL capability was developed by leveraging a decomposed reward Deep Q-Network (drDQN) algorithm. In the defined AMS scenario, drDQN consists of (i) a DQN driven by maximizing mission accomplishment objective and (ii) a DQN driven by minimizing maintenance cost objective. As a result, the proposed strategy can provide real-time aircraft maintenance decisions and explain why those decisions are selected and the trade-offs between the selected action and non-selected ones. Experiments show that the proposed drDQN performs reasonably well, approximating the vanilla DQN with a simpler structure but no explainable ability. Finally, the feasibility of the proposed strategy was demonstrated via a web-based prototype with an intuitive textual and visual user interface.

ABSTRACT. In indoor environments, accurate position estimation of multi-agent systems is challenging due to the lack of Global Navigation Satelite System (GNSS) signals. If the multi-agent system relies upon noisy measurements of the change in position and orientation, the integrated position estimate can drift potentially unboundedly. Magnetic field simultaneous localization and mapping (SLAM) has previously been proposed as a way to compensate for position drift in a single agent. We propose two novel algorithms that allow multiple agents to apply magnetic field SLAM using their own and the other agents' measurements. Our first algorithm is a centralized algorithm that uses all measurements collected by all agents in a single extended Kalman filter. The algorithm simultaneously estimates the agent's position and orientation and the magnetic field norm in a central unit that can communicate with all agents at all times. In other applications, there is no central unit available, and there are communication drop-outs between agents. Our second algorithm is therefore a distributed algorithm for multi-agent magnetic field SLAM, that can be employed even when there is no central unit, and when there are communication failures between agents. We test our algorithms by estimating the position of magnetometers carried by three people in an optical motion capture lab with simulated odometry and simulated communication dropouts between agents. We show that both algorithms are able to compensate for drift in a case where single-agent SLAM is not. We discuss theoretically and experimentally the conditions for the estimate from our distributed algorithm to converge to the estimate from the centralized algorithm. We find that for a communication drop-out rate that is $80\%$, our proposed distributed algorithm on average gives a position estimate with higher accuracy compared to single agent SLAM. Finally, we demonstrate the drift-compensating abilities of our centralized algorithm on a real-life pedestrian localization problem with multiple agents moving inside a building.

ABSTRACT. This paper deals with state estimation of stochastic models with linear state dynamics, continuous or discrete in time. The emphasis is laid on a numerical solution to the state prediction by the time-update step of the grid-point-based point-mass filter (PMF), which is the most computationally demanding part of the PMF algorithm. A novel efficient PMF (ePMF) estimator is proposed, designed, and discussed. By numerical illustrations, it is shown, that the proposed ePMF can lead to a time complexity reduction that exceeds 99.9% without compromising accuracy. The MATLAB® code of the ePMF is released with this paper.

ABSTRACT. Mobile robots are used to disinfect surfaces with C-band ultraviolet light (UV-C). However, the coverage path planning strategy that delivers the right dose of energy, avoiding waste, is still a challenge. This article proposes a lighting sweep strategy for surfaces, based on robot speed, UV-C output power and distance from the surface, to ensure an adequate dose of UV-C light on the surface. In this work, the distance between the scan paths is mathematically defined and, through simulations, its effectiveness is demonstrated. The experimental results show that when using the ideal speed of sweeping, it is possible to disinfect a surface 2.9 times greater than the best case of the static lamp, in the same period of time. The proposed method in this paper is capable of performing surface coverage, delivering at least the target dose of UV-C while reducing energy and time waste.

ABSTRACT. Knowledge about human intention can be beneficial in many disciplines of robotics, such as collaborative manufacturing, prosthetics, or encountered-type haptics. Existing intention estimation approaches are either traditional and rely on hand-crafted features and heuristics, or learning-based and tailored to very specific conditions. This paper attempts to combine the best of both worlds by making recurrent neural networks adaptable to different scenarios. To achieve this, the intention estimation problem is formulated as a probabilistic classification problem and two new data sets with real-world motion and eye-tracking data are presented. Based on this data, three real-time capable classifiers with different features regarding situational awareness and additional outputs are designed and evaluated against two competing approaches. The results show that two out of three classifiers lead to improved or equivalent performance compared to traditional approaches, while good generalization is maintained.

ABSTRACT. A multi-target tracking method requires a data association algorithm to decide which measurements, from a set of many, to use for updating each track. In this paper we will consider a case where the position of multiple marine vessels is measured a coastal region, and the time span between each measurement is large. The vessels of interest are assumed to be sufficiently inert in the sense that the vessels intend to maintain a particular cruising speed while traveling to their destination. Moreover, by assuming that the vessels intend to travel the shortest route possible, we have a foundation to assign conditional probabilities to each measured position. We use the fast marching method (FMM) to compute a time-of-arrival (ToA) map which indicates where the vessel can be at a particular time, provided that the initial speed is known. The ToA map is then evaluated with respect to a Gaussian distribution which provides a probability density function (pdf) used to evaluate the probability that two measurements originate from the same vessel.

ABSTRACT. Unsupervised anomaly detection approaches have been widely accepted in applications for industrial systems. Industrial systems often operate with multiple modes since they work for multiple purposes or under different conditions. In order to deal with the difficulty of anomaly detection due to multiple operating modes, multiple modeling strategies are employed. However, estimating the optimal set of models is a challenging problem due to the lack of supervision and computational burden. In this paper, we propose DeconAnomaly, a deep learning framework to estimate the optimal set of models using transfer learning for unsupervised anomaly detection under a multiple modeling strategy. It reduces computational burden with transfer learning and optimizes the number of models based on a surrogate metric of detection performance. The experimental results show clear advantages of DeconAnomaly.

ABSTRACT. This paper proposes a novel interactive system for exploratory document search in multi-type data sets that employs a data fusion approach. The system, which is designed to visualize different object types collectively and clearly display their semantic proximity, utilizes a co-embedding technique for knowledge fusion of multi-type data and projects the different types of objects onto a common lower-dimensional space. This produces a more informed representation and visualization that shows both in-type and across-type semantic proximity between objects. The system enables users’ exploration of multi-type document data by providing embedding-based scatter plots to visualize semantic relations within and across object types. In addition, based on user relevance feedback of displayed objects, the system provides recommendations of relevant objects. We have demonstrated the effectiveness of the proposed system and the underlying embedding method through comparison experiments of the proposed fusion-based approach.

ABSTRACT. The ensemble Gaussian mixture filter (EnGMF) is a powerful filter for highly non-Gaussian and non-linear models that has practical utility in the case of a small number of samples, and theoretical convergence to full Bayesian inference in the ensemble limit. We aim to increase the utility of the EnGMF by introducing a particle-local notion of covariance into the Gaussian mixture estimate of the prior distribution. We show on a simple bivariate problem that each particle having its own local estimate of the covariance both has nice qualitative and quantitative properties, and significantly improves the estimate of the prior and posterior distributions for all ensemble sizes. We additionally show the utility of the proposed methodology for sequential filtering for the Lorenz '63 equations, achieving a significant reduction in error in the low ensemble size regime.

ABSTRACT. Future autonomous ships will need several redundant positioning systems to navigate reliably. Global Navigation Satellite Systems are highly accurate but they are susceptible to disruptions and intentional jamming. Maritime radars have long range and are robust against bad weather and darkness, but the use for ownship motion estimation has received relatively little attention in the research field. In this work, we present a radar odometry estimation method inspired by advances in visual odometry and simultaneous localization and mapping. The method works on raw radar data in a coastal environment, and combines Kanade-Lucas-Tomashi tracker with a factor graph backend. We test it on data from a large ship with a maritime radar with a range of 19 km. We find that it is robust with only a small drift and no erroneous jumps in the estimate.

ABSTRACT. We present a mapping algorithm to compute large-scale magnetic field maps in indoor environments with approximate Gaussian process (GP) regression. Mapping the spatial variations in the ambient magnetic field can be used for localization algorithms in an indoor area. To compute such a map, GP regression is a suitable tool because it provides predictions of the magnetic field at new locations along with uncertainty quantification. Because full GP regression has a complexity that grows cubically with the number of data points, approximations for GPs have been extensively studied. In this paper, we build on the structured kernel interpolation (SKI) framework, speeding up inference by exploiting efficient Krylov subspace methods. More specifically, we incorporate SKI with derivatives (D-SKI) into the scalar potential model for magnetic field modeling and compute both predictive mean and covariance with a complexity that is linear in the data points. In our simulations, we show that our method achieves better accuracy than current state-of-the-art methods on magnetic field maps with a growing mapping area. In our large-scale experiment, we construct a magnetic field map on more than 40 000 three-dimensional magnetic field measurements in less than two minutes on a standard laptop.

ABSTRACT. Ferromagnetic materials in indoor environments give rise to disturbances in the ambient magnetic field. Maps of these magnetic disturbances can be used for indoor localisation. A Gaussian Process (GP) can be used to learn magnetic field maps using magnetometer measurements and information about the position of the magnetometer. The position of the magnetometer, however, is frequently only approximately known. This negatively affects the quality of the magnetic field map. In this paper, we investigate how an array of magnetometers can be used to improve the quality of the magnetic field map. The position of the array is approximately known, but the relative locations of the magnetometers on the array are known. We include this information in a novel method to make a map of the ambient magnetic field. We study the properties of our method in simulation and show our method improves the map quality. We also demonstrate the efficacy of our novel method with experimental data for the mapping of the magnetic field using an array of 30 magnetometers.