Switching Extended Kalman Filter Bank for Indoor Localization Using Wireless Sensor Networks

Abstract

This paper presents a new filtering algorithm, switching extended Kalman filter bank (SEKFB), for indoor localization using wireless sensor networks. SEKFB overcomes the problem of uncertain process-noise covariance that arises when using the constant-velocity motion model for indoor localization. In the SEKFB algorithm, several extended Kalman filters (EKFs) run in parallel using a set of covariance hypotheses, and the most probable output obtained from the EKFs is selected using Mahalanobis distance evaluation. Simulations demonstrated that the SEKFB can provide accurate and reliable localization without the careful selection of process-noise covariance.

Modeling of Fundus Laser Exposure for Estimating Safe Laser Coagulation Parameters in the Treatment of Diabetic Retinopathy

“…For example, as mentioned earlier, median filtering ensures the elimination of strong impulse noise by brightness replacement with the median value in a certain neighborhood. If the noise is additively distributed over all pixels in the form of additive Gaussian white noise (AGWN), then it is possible to use recurrent Kalman filtering [27]. Consequently, the first task is to highlight different layers in the noisy fundus image.…”

“…Considering very small values of ∆t, it is possible to neglect the diffraction of light [27]. Since this is indeed the case in practice, the model of the heat propagation after laser exposure can be rewritten in the form of Equation ( 3):…”

A Survey of Recent Indoor Localization Scenarios and Methodologies

“…The indoor localization problem can be classified into two principal branches: machine learning methods [18][19][20][21][22][23][24][25][26][27][28][29] and filters-based methods [30][31][32][33][34][35][36][37]. Traditional supervised machine learning methods such as SVM (support vector machine), KNN (k-nearest neighbors), Naive Bayes, and decision tree methods are capable of solving the data extraction, matching, and notably classification issues on the localization problem.…”

“…For the filter-based methods, the particle filter (PF) [34,36,54] and Kalman Filter (KF) [30][31][32][33]55,56]-based methods provide a large number of localization solutions. These filter-based methods are generally consisted of three steps: Prediction, measurement, and assimilation.…”

“…The major difference of particle and Kalman filter is that, Kalman filter is modeled by uni-modal Single Gaussian Model (SGM) with linear functions, whereas particle filter applies factored sampling techniques to indicate the probability density features and to realize dynamic tracking based on Gaussian Mixture Model (GMM). Besides, the extended Kalman filter (EKF) methods [30,32] are more suitable for indoor localization scenario that is non-linear system, e.g., multi-modal Gaussian Model, while PF methods apply Sequential Monte Carlo method to indicate the density function and to predict the current position, which is the reason that EKF is widely applied for accurate indoor localization. EKF is considered as an approximation of Gaussian distribution with nonlinear functions, EKF must remove the second and higher-order derivative terms which contain during the linearization process, but still it is not suitable for strongly non-linear systems.…”