Auto-Regressive Modeling of the Shadowing for RSS Mobile Tracking

Gradient-Based Sequential Markov Chain Monte Carlo for Multitarget Tracking With Correlated Measurements

“…In a single target scenario, the authors in [10] propose to use a sequential Monte-Carlo method, known as particle filter, in order to infer the single target characteristics given the observations. However, this method suffers from intrinsic limitations in high-dimensional systems ( [11], [21]), as the number of samples needs to increase exponentially with the variance of the weights (which is typically a linear function of the state dimension) so as to ensure that not only a single weight will be non-null.…”

“…A few examples of research include [6] which studies the combination of measurement correlation and shrinkage estimation of the covariance matrix for significant performance improvements, but is limited to the static case. In [7]- [10] the measurement correlations are taken into account and refined particle filtering (or Sequential Importance Resampling -SIR) algorithms are implemented. This results in high localization accuracy, however these algorithms inherently suffer from the limitations of the particle filtering approach.…”

Sequential Markov Chain Monte Carlo for multi-target tracking with correlated RSS measurements

“…In a single target scenario, the authors in [10] propose to use a sequential Monte-Carlo method, known as particle filter, in order to infer the single target characteristics given the observations. However, this method suffers from intrinsic limitations in high-dimensional systems [11].…”

“…Now, we compare the proposed SMCMC algorithm with the particle filter that was proposed in [10] in a similar context for single target tracking. The particle filter used in this section is the Sequential Importance Resampling (SIR) [17] in which a resample move strategy after the resampling stage is employed in order to diversify the set of particles [18].…”

“…A few examples of research include [6] which studies the combination of measurement correlation and shrinkage estimation of the covariance matrix for significant performance improvements, but is limited to the static case. In [7][8][9][10] the measurement correlations are taken into account and refined particle filtering (or Sequential Importance Resampling) algorithms are implemented, resulting in high accuracy localization, however they inherently suffer from the limitations of the particle filtering approach, which, although known to be an effective way of solving non-linear problems, performs poorly in high-dimensional state-spaces [11].…”