Browsing by Subject "Distributed"

Now showing 1 - 2 of 2
  • Thumbnail Image
    Item
    Distributed and robust tracking by exploiting set-membership and sarsity.
    (2011-07) Farahmand, Shahrokh
    Target tracking research and development are of major importance and continuously expanding interest to a gamut of traditional and emerging applications, which include radar- and sonar-based systems, surveillance and habitat monitoring using distributed wireless sensors, collision avoidance modules envisioned for modern transportation systems, and mobile robot localization and navigation in static and dynamically changing environments, to name a few. This thesis contributes in several issues pertaining to robustness and distributed operation of modern tracking systems. The first issue addressed relates to measurement model nonlinearity. It turns out that by adopting a grid to describe the surveillance region, the nonlinear measurement model can be cast as a linear one at the cost of increasing state dimensionality. However, by exploiting sparsity of the state in this higher dimension, novel approaches are developed for tracking target signal strengths on a grid (TSSG). In multi-target settings, the proposed sparsity-aware TSSG trackers can bypass the challenge of data association and do not require knowing the number of targets present. To obtain individual target tracks when needed, simple data association techniques are also introduced. Due to the independence of TSSG trackers from the data association stage, association errors do not influence TSSG tracking performance. Mitigating the effect of outliers appearing in the state and measurements is the second topic addressed in this thesis. The proposed robust algorithm referred to as doubly robust smoother (DRS) jointly estimates the outliers alongside with the state. To enable such joint estimation, sparsity in the outlier domain is exploited by regularizing the adopted criterion with the $\ell_1$-norm of the outlier vector. Through novel methods for parameter tuning, DRS is capable of coping with even high levels of outlier contamination. To ensure low-complexity implementation, iterative coordinate descent and method-of-multipliers based solvers are derived for DRS. For settings where the state remains invariant or varies slowly with time, an online robust recursive least-squares (RLS) algorithm referred to as OR-RLS is also derived. Both DRS and OR-RLS are compared against existing robust alternatives and shown to significantly improve the performance. Finally, a consensus-based distributed particle filter referred to as set-membership constrained particle filter (SMC-PF) is introduced for tracking with wireless sensor networks. Consensus-based algorithms possess well-known merits such as robustness to sensor and link failures, scalability, and only local message exchanges with one-hop neighbors. In addition, SMC-PF provides a consensus-based mechanism for data adaptation based on set-membership ideas. SMC-PF outperforms state-of-the-art distributed particle filters in terms of performance and communication complexity.
  • Thumbnail Image
    Item
    ST-Hadoop: A MapReduce Framework for Big Spatio-temporal Data Management
    (2019-05) Alarabi, Louai
    Apache Hadoop, employing the MapReduce programming paradigm, that has been widely accepted as the standard framework for analyzing big data in distributed environments. Unfortunately, this rich framework was not genuinely exploited towards processing large scale spatio-temporal data, especially with the emergence and popularity of applications that create them in large-scale. The huge volumes of spatio-temporal data come from applications, like Taxi fleet in urban computing, Asteroids in astronomy research studies, animal movements in habitat studies, neuron analysis in neuroscience research studies, and contents of social networks (e.g., Twitter or Facebook). Managing space and time are two fundamental characteristics that raised the demand for processing spatio-temporal data created by these applications. Besides the massive size of data, the complexity of shapes and formats associated with these data raised many challenges in managing spatio-temporal data. The goal of the dissertation is centered on establishing a full-fledged big spatio-temporal data management system that serves the need for a wide range of spatio-temporal applications. This involves indexing, querying, and analyzing spatio-temporal data. We propose ST-Hadoop; the first full-fledged open-source system with native support for big spatio-temporal data, available to download http://st-hadoop.cs.umn.edu/. ST- Hadoop injects spatio-temporal data awareness inside the highly popular Hadoop system that is considered state-of-the-art for off-line analysis of big data systems. Considering a distributed environment, we focus on the following: (1) indexing spatio-temporal data and (2) Supporting various fundamental spatio-temporal operations, such as range, kNN, and join (3) Supporting indexing and querying trajectories, which is considered as a special class of spatio-temporal data that require special handling. Throughout this dissertation, we will touch base on the background and related work, motivate for the proposed system, and highlight our contributions.