Aristotle University Of Thessaloniki (AUTH)
Τροποποιημένα φίλτρα Kalman για την εκτίμηση κρυφών καταστάσεων με λογοκριμένες μετρήσεις
Abstract
dc:descriptionThe first chapter of the dissertation introduces the reader in the research field which is under study. The basic principles of stochastic state space models are presented and nonlinear models that are subject to measurement censorship are described. The second chapter presents a brief description of the results and describes the contribution of this dissertation to the relevant research field. In the third chapter, the corrected Tobit Kalman filter is proposed for hidden states estimation with censored multidimensional measurements. The exact probabilities of the measurements belonging to or outside the censored region are calculated, as well as the exact expected value and the variance-covariance matrix of the censored measurement. Next, the adaptive Tobit Kalman filter, with adaptive cesnoring limits at each time step, is proposed to denoise and improve the signal received from the sensor Microsoft Kinect V2. In the fourth chapter, a second way of calculating the moments, up to the third order, for the one-dimensional censored measurements through the moment generating function is presented. In addition, the coloured Tobit Kalman filter is proposed for hidden states estimation when the state space model noises are described by autoregressive models AR(1). The likelihood function of the censored measurements is utilized to estimate the unknown parameters of the AR(1) models. The fifth chapter proposes a novel method for hidden states estimation when one-dimensional measurements are censored. The proposed method, the censored Kalman filter, does not adopt the assumptions of the previous chapters, but calculates in detail the hidden state estimates at each time step, which is considered to be the expected value of the hidden state, given the censored measurement at each time step. Then, the censored Kalman filter is used to track the position and trajectories of pedestrians, as recorded by external cameras.
Degree
thesis:*- Grantor dc:publisher
- Aristotle University Of Thessaloniki (AUTH)
- Year dc:date
- 2020
Author and committee
dc:creator, dc:contributor.*- Authors dc:creator
-
- Loumponias, Kostas
- Λουμπόνιας, Κώστας
Subjects
dc:subject × 14Rights
- Language dc:language
- gre
Identifiers
dc:identifier.*- Identifier
- 10.12681/eadd/55987