Agent-based simulation of smart environment finds its application in studying people’s movement to help the design of a variety of applications such as energy utilization, HAVC control and egress strategy in emergency situation. Traditionally, agent-based simulation is not dynamic data driven, they run offline and do not assimilate real sensor data about the environment. As more and more buildings are equipped with various sensors, it is possible to utilize real time sensor data to inform the simulation. To incorporate the real sensor data into the simulation, we introduce the method of data assimilation. The goal of data assimilation is to provide inference about system state based on the incomplete, ambiguous and uncertain sensor data using a computer model. A typical data assimilation framework consists of a computer model, a series of sensors and a melding scheme. The purpose of this dissertation is to develop a data assimilation framework for agent-based simulation of smart environment. With the developed data assimilation framework, we demonstrate an application of building occupancy estimation which focuses on position estimation using the framework. We build an agent based model to simulate the occupants’ movement s in the building and use this model in the data assimilation framework. The melding scheme we use to incorporate sensor data into the built model is particle filter algorithm. It is a set of statistical method aiming at compute the posterior distribution of the underlying system using a set of samples. It has the benefit that it does not have any assumption about the target distribution and does not require the target system to be written in analytic form .To overcome the high dimensional state space problem as the number of agents increases, we develop a new resampling method named as the component set resampling and evaluate its effectiveness in data assimilation. We also developed a graph-based model for simulating building occupancy. The developed model will be used for carrying out building occupancy estimation with extremely large number of agents in the future.

Particle filters are a class of sequential Monte-Carlo methods which are used commonly when estimating various unknowns of the time-varying signals presented in real time, especially when dealing with nonlinearity and non-Gaussianity in BOT applications. This thesis work is designed to perform one such estimate involving tracking a person using the road information available from an IR surveillance video. In this thesis, a parallel custom hardware is implemented in Altera cyclone IV E FPGA device utilizing SIRF type of particle filter. This implementation has accounted how the algorithmic aspects of this sampling based filter relate to possibilities and constraints in a hardware implementation. Using 100MHz clock frequency, the synthesised hardware design can process almost 50 Mparticles/s. Thus, this implementation has resulted in tracking the target, which is defined by a 5-dimensional state variable, using the noisy measurements available from the sensor.

In this master thesis, image processing methods and methods for statistical modeling of motion are presented. First, description methods of image processing, such as background subtraction method used for object detection, are presented. Next, description of morphological operations, such as dilatation and erosion, is done. Finally, methods for statistical modeling, such as Kalman filter and particle filters, are shown.

No description available.

Understanding machine vision can certainly improve our understanding of artificial intelligence as vision happens to be one of the basic intellectual activities of living beings. Since the notion of computation unifies the concept of a machine, computer vision can be understood as an application of modern approaches for achieving artificial intelligence, like machine learning and cognitive psychology. Computer vision mainly involves processing of different types of sensor data resulting in "perception of machines". Perception of machines plays a very important role in several artificial intelligence applications with sensors. There are numerous practical situations where we acquire sensor data for e.g. from mobile robots, security cameras, service and recreational robots. Making sense of this sensor data is very important so that we have increased automation in using the data. Tools from image processing, shape analysis and probabilistic inferences i.e. learning theory form the artillery for current generation of computer vision researchers. In my thesis I will address some of the most annoying components of two important open problems viz. object recognition and autonomous navigation that remain central in robotic, or in other words computational, intelligence. These problems are concerned with inducing computers, the abilities to recognize and navigate similar to those of humans. Object boundaries are very useful descriptors for recognizing objects. Extracting boundaries from real images has been a notoriously open problem for several decades in the vision community. In the first part I will present novel techniques for extracting object boundaries. The techniques are based on practically successful state-of-the-art Bayesian filtering framework, well founded geometric properties relating boundaries and skeletons and robust high-level shape analyses Acquiring global maps of the environments is crucial for robots to localize and be able to navigate autonomously. Though there has been a lot of progress in achieving autonomous mobility, for e.g. as in DARPA grand-challenges of 2005 and 2007, the mapping problem itself remains to be unsolved which is essential for robust autonomy in hard cases like rescue arenas and collaborative exploration. In the second part I will present techniques for merging maps acquired by multiple and single robots. We developed physics-based energy minimization techniques and also shape based techniques for scalable merging of maps. Our shape based techniques are a product of combining of high-level vision techniques that exploit similarities among maps and strong statistical methods that can handle uncertainties in Bayesian sense. / Computer and Information Science

Energy efficiency is a critical feature of wireless sensor networks (WSNs), because sensor nodes run on batteries that are generally difficult to recharge once deployed. For target tracking---one of the most important WSN application types---energy efficiency needs to be considered in various forms and shapes, such as idle listening, trajectory estimation, and data propagation. In this dissertation, we study three correlated problems on energy efficient target tracking in WSNs: sleep scheduling, particle filtering, and constrained flooding. We develop a Target Prediction and Sleep Scheduling protocol (TPSS) to improve energy efficiency for idle listening. We start with designing a target prediction method based on both kinematics and probability. Based on target prediction and proactive wake-up, TPSS precisely selects the nodes to awaken and reduces their active time, so as to enhance energy efficiency with limited tracking performance loss. In addition, we expand Sleep Scheduling to Multiple Target Tracking (SSMTT), and further reduce the energy consumption by leveraging the redundant alarm messages of interfering targets. Our simulation-based experimental studies show that compared to existing protocols such as Circle scheme and MCTA, TPSS and SSMTT introduce an improvement of 25% ~ 45% on energy efficiency, at the expense of only 5% ~ 15% increase on the detection delay. Particle Filtering is one of the most widely used Bayesian estimation methods, when target tracking is considered as a dynamic state estimation problem for trajectory estimation. However, the significant computational and communication complexity prohibits its application in WSNs. We design two particle filters (PFs)---Vector space based Particle Filter (VPF) and Completely Distributed Particle Filter (CDPF)---to improve energy efficiency of PFs by reducing the number of particles and the communication cost. Our experimental evaluations show that even though VPF incurs 34% more estimation error than RPF, and CDPF incurs a similar estimation error to SDPF, they significantly improve the energy efficiency by as much as 68% and 90% respectively. For data propagation, we present a Constrained Flooding protocol (CFlood) to enhance energy efficiency by increasing the deadline satisfaction ratio per unit energy consumption of time-sensitive packets. CFlood improves real-time performance by flooding, but effectively constrains energy consumption by controlling the scale of flooding---i.e., flooding only when necessary. If unicasting meets the distributed sub-deadline at a hop, CFlood aborts further flooding even after flooding has occurred in the current hop. Our simulation-based experimental studies show that CFlood achieves higher deadline satisfaction ratio per unit energy consumption by as much as 197%, 346%, and 20% than existing multipath forwarding protocols, namely, Mint Routing, MCMP and DFP respectively, especially in sparsely deployed or unreliable sensor network environments. To verify the performance and efficiency of the dissertation's solutions, we developed a prototype implementation based on TelosB motes and TinyOS version 2.1.1. In the field experiments, we compared TPSS, VPF, CDPF, and CFlood algorithms/protocols to their respective competing efforts. Our implementation measurements not only verified the rationality and feasibility of the proposed solutions for target tracking in WSNs, but also strengthened the observations on their efficiency from the simulation. / Ph. D.

In recent years, unmanned aerial vehicles (UAVs) have got a boost in their applications in civilian areas like aerial photography, agriculture, communication, etc. An increasing research effort is being exerted to develop sophisticated trajectory prediction methods for UAVs for collision detection and trajectory planning. The existing techniques suffer from problems such as inadequate uncertainty quantification of predicted trajectories. This work adopts particle filters together with Löwner-John ellipsoid to approximate the highest posterior density region for trajectory prediction and uncertainty quantification. The particle filter is tuned and tested on real-world and simulated data sets and compared with the Kalman filter. A parallel computing approach for particle filter is further proposed. This parallel implementation makes the particle filter faster and more suitable for real-time online applications. / Master of Science / In recent years, unmanned aerial vehicles (UAVs) have got a boost in their applications in civilian areas like aerial photography, agriculture, communication, etc. Over the coming years, the number of UAVs will increase rapidly. As a result, the risk of mid-air collisions grows, leading to property damages and possible loss of life if a UAV collides with manned aircraft. An increasing research effort has been made to develop sophisticated trajectory prediction methods for UAVs for collision detection and trajectory planning. The existing techniques suffer from problems such as inadequate uncertainty quantification of predicted trajectories. This work adopts particle filters, a Bayesian inferencing technique for trajectory prediction. The use of minimum volume enclosing ellipsoid to approximate the highest posterior density region for prediction uncertainty quantification is also investigated. The particle filter is tuned and tested on real-world and simulated data sets and compared with the Kalman filter. A parallel computing approach for particle filter is further proposed. This parallel implementation makes the particle filter faster and more suitable for real-time online applications.

Title: Sequential Monte Carlo Methods Author: David Coufal Department: Department of Probability and Mathematical Statistics Supervisor: prof. RNDr. Viktor Beneš, DrSc. Abstract: The thesis summarizes theoretical foundations of sequential Monte Carlo methods with a focus on the application in the area of particle filters; and basic results from the theory of nonparametric kernel density estimation. The summary creates the basis for investigation of application of kernel meth- ods for approximation of densities of distributions generated by particle filters. The main results of the work are the proof of convergence of kernel estimates to related theoretical densities and the specification of the development of approx- imation error with respect to time evolution of a filter. The work is completed by an experimental part demonstrating the work of presented algorithms by simulations in the MATLABR⃝ computational environment. Keywords: sequential Monte Carlo methods, particle filters, nonparametric kernel estimates

The world went through weeks of financial turbulence in stock markets and investors were overcome by fears fuelled by more bad news, while countries continued their attempts to calm the markets with more injection of funds. By these very disturbed times, even if traders hope extreme risk aversion has passed, an investor would like predict the future of the market in order to protect his portfolio and a speculator would like to optimize his tradings. This thesis describes the design of numerical models and algorithms for the forecasting of financial time series, for speculation on a short time interval. To this aim, we will use two models: - " Price Forecasting Model " forecasts the behavior of an asset for an interval of three hours. This model is based on Functional Clustering and smoothing by cubic-splines in the training phase to build local Neural models, and Functional Classification for generalization, - " Model of Trading " forecasts the First Stopping time, when an asset crosses for the first time a threshold defined by the trader. This model combines a Price Forecasting Model for the prediction of market trend, and a Trading Recommendation for prediction of the first stopping time. We use an auto-adaptive Dynamic State Space Model, with Particle Filters and Kalman-Bucy Filters for parameter estimation.

Sequential Monte Carlo (SMC) and Markov chain Monte Carlo (MCMC) methods provide computational tools for systematic inference and learning in complex dynamical systems, such as nonlinear and non-Gaussian state-space models. This thesis builds upon several methodological advances within these classes of Monte Carlo methods.Particular emphasis is placed on the combination of SMC and MCMC in so called particle MCMC algorithms. These algorithms rely on SMC for generating samples from the often highly autocorrelated state-trajectory. A specific particle MCMC algorithm, referred to as particle Gibbs with ancestor sampling (PGAS), is suggested. By making use of backward sampling ideas, albeit implemented in a forward-only fashion, PGAS enjoys good mixing even when using seemingly few particles in the underlying SMC sampler. This results in a computationally competitive particle MCMC algorithm. As illustrated in this thesis, PGAS is a useful tool for both Bayesian and frequentistic parameter inference as well as for state smoothing. The PGAS sampler is successfully applied to the classical problem of Wiener system identification, and it is also used for inference in the challenging class of non-Markovian latent variable models.Many nonlinear models encountered in practice contain some tractable substructure. As a second problem considered in this thesis, we develop Monte Carlo methods capable of exploiting such substructures to obtain more accurate estimators than what is provided otherwise. For the filtering problem, this can be done by using the well known Rao-Blackwellized particle filter (RBPF). The RBPF is analysed in terms of asymptotic variance, resulting in an expression for the performance gain offered by Rao-Blackwellization. Furthermore, a Rao-Blackwellized particle smoother is derived, capable of addressing the smoothing problem in so called mixed linear/nonlinear state-space models. The idea of Rao-Blackwellization is also used to develop an online algorithm for Bayesian parameter inference in nonlinear state-space models with affine parameter dependencies. / CNDM / CADICS