In this dissertation, joint detection-state estimation and secure signal processing are studied. Detection and state estimation are two important research topics in surveillance systems. The detection problems investigated in this dissertation include object detection and fault detection. The goal of object detection is to determine the presence or absence of an object under measurement uncertainty. The aim of fault detection is to determine whether or not the measurements are provided by faulty sensors. State estimation is to estimate the states of moving objects from measurements with random measurement noise or disturbance, which typically consist of their positions and velocities over time. Detection and state estimation are typically implemented separately and state estimation is usually performed after the decision is made. In this two-stage approach, missed detection and false alarms in detection stage decrease accuracy of state estimation. In this dissertation, several joint detection and state estimation algorithms are proposed. Secure signal processing is indispensable in dynamic systems especially when an adversary exists. In this dissertation, the developed joint fault detection and state estimation approach is used to detect attacks launched by an adversary on the system and improve state estimation accuracy. The security problem in satellite communication systems is studied and a minimax anti-jammer is designed in a frequency hopping spread spectrum (FHSS)/quadrature phase-shift keying (QPSK) satellite communication system.

The impact of false information injection is investigated for linear dynamic systems with multiple sensors. First, it is assumed that the system is unaware of the existence of false information and the adversary is trying to maximize the negative effect of the false information on Kalman filter's estimation performance under a power constraint. The false information attack under different conditions is mathematically characterized. For the adversary, many closed-form results for the optimal attack strategies that maximize the Kalman filter's estimation error are theoretically derived. It is shown that by choosing the optimal correlation coefficients among the false information and allocating power optimally among sensors, the adversary could significantly increase the Kalman filter's estimation errors. In order to detect the false information injected by an adversary, we investigate the strategies for the Bayesian estimator to detect the false information and defend itself from such attacks. We assume that the adversary attacks the system with certain probability, and that he/she adopts the worst possible strategy that maximizes the mean squared error (MSE) if the attack is undetected. An optimal Bayesian detector is designed which minimizes the average system estimation error instead of minimizing the probability of detection error, as a conventional Bayesian detector typically does. The case that the adversary attacks the system continuously is also studied. In this case, sparse attack strategies in multi-sensor dynamic systems are investigated from the adversary's point of view. It is assumed that the defender can perfectly detect and remove the sensors once they are corrupted by false information injected by an adversary. The adversary's goal is to maximize the covariance matrix of the system state estimate by the end of attack period under the constraint that the adversary can only attack the system a few times over the sensor and over the time, which leads to an integer programming problem. In order to overcome the prohibitive complexity of the exhaustive search, polynomial-time algorithms, such as greedy search and dynamic programming, are proposed to find the suboptimal attack strategies. As for greedy search, it starts with an empty set， and one sensor is added at each iteration, whose elimination will lead to the maximum system estimation error. The process terminates when the cardinality of the active set reaches to the sparsity constraint. Greedy search based approaches such as sequential forward selection (SFS), sequential backward selection (SBS), and simplex improved sequential forward selection (SFS-SS) are discussed and corresponding attack strategies are provided. Dynamic programming is also used in obtaining the sub-optimal attack strategy. The validity of dynamic programming lies on a straightforward but important nature of dynamic state estimation systems: the credibility of the state estimate at current step is in accordance with that at previous step. The problem of false information attack on and the Kalman filter's defense of state estimation in dynamic multi-sensor systems is also investigated from a game theoretic perspective. The relationship between the Kalman filter and the adversary can be regarded as a two-person zero-sum game. The condition under which both sides of the game will reach a Nash equilibrium is investigated.

Existing estimations of implicit inflation target are primarily based on the assumption of parameter stability over time horizon. This work relaxes this assumption and proposes alternative framework based on time-varying parameter model. We aim on behaviour of European Central Bank in order to compare its official proclamations of price stability levels with our implicit estimations. We will also examine how two pillar strategy of European Central Bank is practically used.

The extended Kalman lter (EKF) has a long history in the field of non-linear tracking. More recently, statistically-based estimators have emerged that avoid the need for a deterministic linearisation process. The Unscented Kalman filter (UKF) is one such technique that has been shown to perform favourably for some non-linear systems when compared to an EKF implementation, both in terms of accuracy and robustness. In this Thesis, the UKF is applied to a high energy physics particle tracking problem where currently the EKF is being implemented. The effects of measurement redundancy are investigated to determine improvements in accuracy of particle track reconstruction. The relationship between measurement redundancy and relative observability is also investigated through an experimental and theoretical analysis. Smoothing (backward filtering), in the high energy physics experiments, is implementedusing the Rauch Tung Striebel (RTS) smoother with the EKF , however, in Unscented Kalman filter algorithms, the Jacobian matrices required by the RTS method, are not available. The Unscented Rauch Tung Striebel (URTS) smoother addresses this problem by avoiding the use of Jacobian matrices but is not effi cient for large dimensional systems such as high energy physics experiments. A technique is implemented in the RTS smoother to make it suitable for the UKF. The method is given the name the Jacobian Equivalent Rauch Tung Striebel (JE-RTS) smoother. The implementation of this method is quite straight forward when the UKF is used as an estimator.

Doctor of Philosophy / Department of Electrical and Computer Engineering / Balasubramaniam Natarajan / A cyber physical system (CPS) is an intelligent integration of computation and communication infrastructure for monitoring and/or control of an underlying physical system. In this dissertation, we consider a specific class of CPS architectures where state of the system is spatially distributed in physical space. Examples that fit this category of CPS include, smart distribution gird, smart highway/transportation network etc. We study state estimation and control process in such systems where, (1) multiple sensors and actuators are arbitrarily deployed to jointly sense and control the system; (2) sensors directly communicate their observations to a central estimation and control unit (ECU) over communication links; and, (3) the ECU, on computing the control action, communicates control actions to actuators over communication links. Since communication links are susceptible to random failures, the overall estimation and control process is subjected to: (1) partial observation updates in estimation process; and (2) partial actuator actions in control process. We analyze stochastic stability of estimation and control process, in this scenario by establishing the conditions under which estimation accuracy and deviation from desired state trajectory is bounded. Our key contribution is the derivation of a new fundamental result on bounds for critical probabilities of individual communication link failure to maintain stability of overall system. The overall analysis illustrates that there is trade-off between stability of estimation and control process and quality of underlying communication network. In order to demonstrate practical implication of our work, we also present a case study in smart distribution grid as a system example of spatially distributed CPSs. Voltage/VAR support via distributed generators is studied in a stochastic nonlinear control framework.

[pt] Este trabalho aborda o problema da Localização e Mapeamento Simultâneos em ambientes estruturados, utilizando um robô móvel equipado com sonares, bússola eletrônica e encoders. Na modelagem sugerida há a construção do mapa do ambiente e a localização do robô de forma interativa. O método proposto, denominado de LMS-H (Localização e Mapeamento Simultâneos - Híbrido), faz uso de duas formas de representação do ambiente: Mapa de Ocupação em Grade e Representação Contínua. O Mapa de Ocupação em Grade divide o ambiente em pequenas partes iguais, classificando-as em ocupadas ou vazias. A Representação Contínua utiliza retas para representar os planos detectados no ambiente, formando um mapa em duas dimensões e cada reta do mapa é considerada um marco. Sempre que um plano é novamente detectado pelo robô a reta correspondente a ele é recalculada com os novos pontos obtidos e a posição do robô é atualizada via Filtro de Kalman Estendido. A eficácia do método foi comprovada através de seis estudos de caso: três em ambientes virtuais e três em ambientes reais. Os estudos de casos em ambientes reais foram realizados utilizando-se um protótipo feito sob a plataforma LEGO Mindstorms. Os resultados obtidos comprovaram a eficácia do método proposto. / [en] This work addresses the problem of Simultaneous Localization and Mapping in structured environments using a mobile robot equipped with sonar, electronic compass and encoders. In the proposed modeling there are the construction of the environment map and the robot localization interactively. The proposed method, called H-SLAM (Hybrid - Simultaneous Localization and Mapping), makes use kinds of environment representation: Occupancy Grid Map and Continuous Representation. The Occupancy Grid Map divides the environment into small equal parts, and classifies it as occupied or empty. The Continuous Representation uses lines to represent detected planes in the environment, forming a two-dimensional map. Each line of the map is considered a landmark. Every time a plan is redetected by the robot the corresponding line to it is rebuild with the new points obtained and the robot s position is updated through Extended Kalman Filter. The model effectiveness was proved with computer simulations in three virtual environments. Using a prototype developed with LEGO Mindstorms platform three other experiments were also performed in real environments. The results demonstrated the effectiveness of the proposed method.

In this thesis we present a detect-tracking algorithm (see figure 3.1) that combines the detection robustness of static region proposal based object detectors, like the faster region convolutional neural network (R-CNN) and the region-based fully convolutional networks (R-FCN) model, with the tracking prediction strength of extended Kalman filters, by using, what we have called, a translating and non-rigid user input region of interest (RoI-) mapping. This so-called RoI-mapping maps a region, which includes the object that one is interested in tracking, to a featureless three-channeled image. The detection part of our proposed algorithm is then performed on the image that includes only the RoI features (see figure 3.2). After the detection step, our model re-maps the RoI features to the original frame, and translates the RoI to the center of the prediction. If no prediction occurs, our proposed model integrates a temporal dependence through a Kalman filter as a predictor; this filter is continuously corrected when detections do occur. To train the region proposal based object detectors that we integrate into our detect-tracking model, we used TensorFlow®’s object detection api, with a random search hyperparameter tuning, where we fine-tuned, all models from TensorFlow® slim base network classification checkpoints. The trained region proposal based object detectors used the inception V2 base network for the faster R-CNN model and the R-FCN model, while the inception V3 base network only was applied to the faster R-CNN model. This was made to compare the two base networks and their corresponding affects on the detection models. In addition to the deep learning part of this thesis, for the implementation part of our detect-tracking model, like for the extended Kalman filter, we used Python and OpenCV® . The results show that, with a stationary camera reference frame, our proposed detect-tracking algorithm, combined with region proposal based object detectors on images of size 414 × 740 × 3, can detect and track a small object in real-time, like a tennis ball, moving along a horizontal trajectory with an average velocity v ≈ 50 km/h at a distance d = 25 m, with a combined detect-tracking frequency of about 13 to 14 Hz. The largest measured state error between the actual state and the predicted state from the Kalman filter, at the aforementioned horizontal velocity, have been measured to be a maximum of 10-15 pixels, see table 5.1, but in certain frames where many detections occur this error has been shown to be much smaller (3-5 pixels). Additionally, our combined detect-tracking model has also been shown to be able to handle obstacles and two learnable features that overlap, thanks to the integrated extended Kalman filter. Lastly, our detect-tracking model also was applied on a set of infra-red images, where the goal was to detect and track a moving truck moving along a semi-horizontal path. Our results show that a faster R-CNN inception V2 model was able to extract features from a sequence of infra-red frames, and that our proposed RoI-mapping method worked relatively well at detecting only one truck in a short test-sequence (see figure 5.22).

This thesis studies the possibility to replace the global navigation satellite system (GNSS) with a phased array radio system (PARS) for positioning and navigation of an unmanned aerial vehicle (UAV). With the increase of UAVs in both civilian and military applications, the need for a robust and accurate navigation solution has increased. The GNSS is the main solution of today for UAV navigation and positioning. However, the GNSS can be disturbed by malicious sources, the signal can either be blocked by jamming or modified to give the wrong position by spoofing. Studies have been conducted to replace or support the GNSS measurements with other drift free measurements, e.g. camera or radar systems. The position measurements from PARS alone is shown not to provide sufficient quality for the application in mind. The PARS measurements are affected by noise and outliers. Reflections from the ground makes the PARS elevation measurements unusable for this application. A root mean square error (RMSE) accuracy of 10 m for a shorter flight and 198 m for a longer flight are achieved in the horizontal plane. The decrease in accuracy for the longer flight is assumed to come from a range bias that increases with distance due to the flat earth approximation used as the navigation frame. Positioning based on PARS aided with a filter and other GNSS independent sensors is shown to reduce the noise and remove the outliers. Five filters are derived and evaluated: a constant velocity extended Kalman filter (EKF), an inertial measurement unit (IMU) aided EKF, an IMU and barometer aided EKF, a converted measurements Kalman filter (CMKF) and a stationary Kalman filter (KF). The IMU and barometer aided EKF performed the best results with a RMSE of 8 m for a shorter flight and 106 m for a longer flight. The noise is significantly reduced compared to the standalone PARS measurements. The conclusion is that PARS can be used as a redundancy system with the IMU and barometer aided EKF. If the EKF algorithm is too computational demanding, the simpler stationary KF can be motivated since the accuracy is similar to the EKF. The GNSS solution should still be used as the primary navigation solution as it is more accurate.

Este trabalho apresenta através de simulação um estudo comparativo de estimadores de velocidade de motor de indução trifásico por meio de observadores de estado e da técnica do filtro de Kalman. É realizada uma análise comparativa de desempenho das estratégias de estimação determinísticas e estocásticas, com observadores adaptativos e estimadores baseados na teoria do filtro de Kalman estendido, respectivamente. A realização do trabalho visa a constatação dos procedimentos de elaboração, de operação e de aplicação destas técnicas de estimação usando um exemplo real com fins de ilustrar o ensino de controle e acionamento de máquinas elétricas. As simulações foram realizadas através do Matlab/Simulink com a utilização das ferramentas do Power System Blockset (PSB) e o algoritmo dos estimadores é escrito em programa Matlab e executado por uma função S-Function. Os resultados de simulação demonstram a eficiência de cada um dos estimadores propostos, no que se refere ao comportamento transitório, robustez a ruídos e variações nos parâmetros do motor. / This works presents through of the simulation a comparative study of the sensorless of speed estimation of induction three-phase motor using state observer and Kalman filter. A comparative analysis of the performance of the deterministic and stochastic estimation strategies using adaptive observers and estimators based on extended Kalman filter was realized. The work aims to verify the procedure of the elaboration, operation and application of such estimation techniques using a real example to illustrate the teaching of the control and driving of electric machines. The simulations where performed using Matlab/Simulink with Power System Blockset (PSB) toolboxes and the estimators are programmed as S-Function Matlab. The results indicate the effectiveness of the proposed estimators, according to the transient behavior, robustness to noise and ability to handle parametric variations.

Estudar as propriedades físicas do solo envolve conhecer a umidade, o transporte de água e solutos, a densidade, a identificação da porosidade, o que é essencial para o crescimento de raízes das plantas. Para esses estudos, a tomografia de raios X tem se mostrado uma técnica útil. As imagens tomográficas são obtidas através de projeções (sinais) que são reconstruídos com algoritmos adequados. No processo de aquisição dessas projeções, podem surgir ruídos provenientes de diferentes fontes. O sinal tomográfico apresenta ruídos que possuem uma distribuição de Poisson gerada pela contagem de fótons, bem como o detector de fótons é influenciado por uma presença de ruído eletrônico com uma distribuição Gaussiana. Essas diferentes distribuições podem ser mapeadas com transformadas não lineares específicas que alteram uma distribuição Gaussiana para outros tipos de distribuições, como a de transformada de Anscombe (Poisson) ou transformada de Box-Muller (Uniforme), mas são aproximações que apresentam erros acumulativos. As transformadas podem ser então mapeadas por um sistema de redes neurais, o que garante um melhor resultado com o filtro de Kalman não linear em que os pesos da rede e as medidas das projeções são estimados em conjunto. Este trabalho apresenta uma nova solução com filtragem de Kalman descentralizada utilizando redes neurais artificiais embarcada em uma arquitetura reconfigurável com o intuito de obter se um valor ótimo de melhoria na relação Sinal/Ruído de projeções tomográficas e consequentemente nas imagens reconstruídas proporcionando melhorias para os métodos de análise dos físicos de solos agrícolas. / To study the physical properties of soil moisture involves knowing the transport of water and solutes, density, porosity identification, which is essential for the growth of plant roots. For these studies, X-ray tomography has been shown to be a useful technique. The tomographic images are obtained through projections (signals) that are reconstructed with appropriate algorithms. In the process of acquiring these projections, noise can arise from different sources. The tomographic signal is noisy which have a Poisson distribution generated by photon counting, and the photon detector is influenced by a presence of electronic noise with a Gaussian distribution. These different distributions can be mapped to specific nonlinear transformed altering a Gaussian distribution for other types of distributions, such as the Anscombe transform (Poisson) or Box-Muller transform (Uniform), but are approximations that have cumulative errors. Transforms can then be mapped by a neural network system, which ensures a better result with nonlinear Kalman filter in which the network weights and measures of the projections are estimated together. This work presents a new solution to the unscented Kalman filtering using artificial neural networks embedded in a reconfigurable architecture in order to obtain an optimum value of improvement in S/N ratio of tomographic projections and consequently the images reconstructed by providing improvements for the methods of physical parameters of the agricultural soils.