added section recursive state estimation, and 3/4 of related work
This commit is contained in:
Toni
@@ -18,16 +18,20 @@ On the other hand, fixed-interval smoothing requires all observations until time
|
||||
The origin of MC smoothing can be traced back to Genshiro Kitagawa.
|
||||
In his work \cite{kitagawa1996monte} he presented the simplest form of smoothing as an extension to the particle filter.
|
||||
This algorithm is often called the filter-smoother since it runs online and a smoothing is provided while filtering.
|
||||
This approach can produce an accurate approximation of the filtering posterior $p(\vec{q}_{t} \mid \vec{o}_{1:t})$ with computational complexity of only $\mathcal{O}(N)$.
|
||||
\commentByFrank{kleines n?}
|
||||
However, it gives a poor representation of previous states \cite{Doucet11:ATO}.
|
||||
\commentByFrank{wenn noch platz, einen satz mehr dazu warum es schlecht ist?}
|
||||
This approach uses the particle filter steps to update weighted paths $\{(\vec{q}_{1:t}^i , w^i_t)\}^N_{i=1}$, producing an accurate approximation of the filtering posterior $p(\vec{q}_{t} \mid \vec{o}_{1:t})$ with a computational complexity of only $\mathcal{O}(N)$.
|
||||
However, it gives a poor representation of previous states due a monotonic decrease of distinct particles caused by resampling of each weighted path \cite{Doucet11:ATO}.
|
||||
Based on this, more advanced methods like the forward-backward smoother \cite{doucet2000} and backward simulation \cite{Godsill04:MCS} were developed.
|
||||
Both methods are running backwards in time to reweight a set of particles recursively by using future observations.
|
||||
Algorithmic details will be shown in section \ref{sec:smoothing}.
|
||||
|
||||
%wo werden diese eingesetzt, paar beispiele. offline, online
|
||||
In recent years, smoothing gets attention mainly in the field of computer vision and ... Here, ...
|
||||
In recent years, smoothing gets attention mainly in other areas as indoor localisation.
|
||||
The early work of \cite{isard1998smoothing} demonstrates the possibilities of smoothing for visual tracking.
|
||||
They used a combination of the CONDENSATION particle filter with a forward-backward smoother.
|
||||
Based on this pioneering approach, many different solutions for visual and multi-target tracking have been developed \cite{Perez2004}.
|
||||
For example, in \cite{Platzer:2008} a particle smoother is used to reduce multimodalities in a blood flow simulation for human vessels. Or \cite{}
|
||||
|
||||
|
||||
|
||||
Nevertheless, their are some promising approach for indoor localisation systems as well. For example ...
|
||||
|
||||
|
||||
@@ -1,11 +1,8 @@
|
||||
\section{Recursive State Estimation}
|
||||
|
||||
\commentByFrank{schon mal kopiert, dass es da ist.}
|
||||
\commentByFrank{die neue activity in die observation eingebaut}
|
||||
\commentByFrank{magst du hier auch gleich smoothing ansprechen? denke es würde sinn machen weils ja zum kompletten systemablauf gehört und den hatten wir hier ja immer drin. oder was meinst du?}
|
||||
|
||||
We consider indoor localisation as a time-sequential, non-linear and non-Gaussian state estimation problem.
|
||||
Using a recursive Bayes filter that satisfies the Markov property, the posterior distribution at time $t$ can be written as
|
||||
As mentioned before, most smoothing methods require a preceding filtering.
|
||||
In our previous work \cite{Ebner-16}, we consider indoor localisation as a time-sequential, non-linear and non-Gaussian state estimation problem.
|
||||
Therefore, a Bayes filter that satisfies the Markov property is used to calculate the posterior, which is given by
|
||||
%
|
||||
\begin{equation}
|
||||
\arraycolsep=1.2pt
|
||||
@@ -13,40 +10,33 @@
|
||||
&p(\mStateVec_{t} \mid \mObsVec_{1:t}) \propto\\
|
||||
&\underbrace{p(\mObsVec_{t} \mid \mStateVec_{t})}_{\text{evaluation}}
|
||||
\int \underbrace{p(\mStateVec_{t} \mid \mStateVec_{t-1}, \mObsVec_{t-1})}_{\text{transition}}
|
||||
\underbrace{p(\mStateVec_{t-1} \mid \mObsVec_{1:t-1})d\vec{q}_{t-1}}_{\text{recursion}} \enspace,
|
||||
\underbrace{p(\mStateVec_{t-1} \mid \mObsVec_{1:t-1})d\vec{q}_{t-1}}_{\text{recursion}} \enspace.
|
||||
\end{array}
|
||||
\label{equ:bayesInt}
|
||||
\end{equation}
|
||||
%
|
||||
where $\mObsVec_{1:t} = \mObsVec_{1}, \mObsVec_{1}, ..., \mObsVec_{t}$ is a series of observations up to time $t$.
|
||||
The hidden state $\mStateVec$ is given by
|
||||
Here, the previous observation $\mObsVec_{t-1}$ is included into the state transition \cite{Koeping14-PSA}.
|
||||
For approximating eq. \eqref{equ:bayesInt} by means of MC methods, the transition is used as proposal distribution, also known as CONDENSATION algorithm \cite{isard1998smoothing}.
|
||||
|
||||
In context of indoor localisation, the hidden state $\mStateVec$ is defined as follows:
|
||||
\begin{equation}
|
||||
\mStateVec = (x, y, z, \mStateHeading, \mStatePressure),\enskip
|
||||
x, y, z, \mStateHeading, \mStatePressure \in \R \enspace,
|
||||
\end{equation}
|
||||
%
|
||||
where $x, y, z$ represent the position in 3D space, $\mStateHeading$ the user's heading and $\mStatePressure$ the
|
||||
relative atmospheric pressure prediction in hectopascal (hPa).
|
||||
\commentByFrank{hier einfach kuerzen und aufs fusion paper verweisen? auch wenn das noch ned durch ist?}
|
||||
The recursive part of the density estimation contains all information up to time $t-1$.
|
||||
Furthermore, the state transition $p(\mStateVec_{t} \mid \mStateVec_{t-1}, \mObsVec_{t-1})$ models the pedestrian's movement as described in section \ref{sec:trans}.
|
||||
%It should be noted, that we also include the current observation $\mObsVec_{t}$ in it.
|
||||
As proven in \cite{Koeping14-PSA}, we may include the observation $\mObsVec_{t-1}$ into the state transition.
|
||||
|
||||
Containing all relevant sensor measurements to evaluate the current state, the observation vector is defined as follows:
|
||||
where $x, y, z$ represent the position in 3D space, $\mStateHeading$ the user's heading and $\mStatePressure$ the relative atmospheric pressure prediction in hectopascal (hPa). Further, the observation is given by
|
||||
%
|
||||
\begin{equation}
|
||||
\mObsVec = (\mRssiVec_\text{wifi}, \mRssiVec_\text{ib}, \mObsHeading, \mObsSteps, \mObsPressure, \mObsActivity) \enspace,
|
||||
\mObsVec = (\mRssiVec_\text{wifi}, \mRssiVec_\text{ib}, \mObsHeading, \mObsSteps, \mObsPressure, x) \enspace,
|
||||
\end{equation}
|
||||
%
|
||||
where $\mRssiVec_\text{wifi}$ and $\mRssiVec_\text{ib}$ contain the measurements of all nearby \docAP{}s (\docAPshort{})
|
||||
and \docIBeacon{}s, respectively. $\mObsHeading$ and $\mObsSteps$ describe the relative angular change and the number
|
||||
of steps detected for the pedestrian. $\mObsPressure$ is the relative barometric pressure with respect to a fixed reference.
|
||||
Finally, $\mObsActivity$ contains the activity, currently estimated for the pedestrian, which is one of: unknown, standing, walking or
|
||||
walking stairs.
|
||||
%For further information on how to incorporate such highly different sensor types,
|
||||
%one should refer to the process of probabilistic sensor fusion \cite{Khaleghi2013}.
|
||||
By assuming statistical independence of all sensors, the probability density of the state evaluation is given by
|
||||
covering all relevant sensor measurements.
|
||||
Here, $\mRssiVec_\text{wifi}$ and $\mRssiVec_\text{ib}$ contain the measurements of all nearby \docAP{}s (\docAPshort{}) and \docIBeacon{}s, respectively.
|
||||
$\mObsHeading$ and $\mObsSteps$ describe the relative angular change and the number of steps detected for the pedestrian.
|
||||
$\mObsPressure$ is the relative barometric pressure with respect to a fixed reference.
|
||||
Finally, $x$ contains the activity, currently estimated for the pedestrian, which is one of: unknown, standing, walking or walking stairs.
|
||||
|
||||
The probability density of the state evaluation is given by
|
||||
%
|
||||
\begin{equation}
|
||||
%\begin{split}
|
||||
@@ -54,23 +44,16 @@
|
||||
p(\vec{o}_t \mid \vec{q}_t)_\text{baro}
|
||||
\,p(\vec{o}_t \mid \vec{q}_t)_\text{ib}
|
||||
\,p(\vec{o}_t \mid \vec{q}_t)_\text{wifi}
|
||||
\enspace.
|
||||
\enspace
|
||||
%\end{split}
|
||||
\label{eq:evalBayes}
|
||||
\end{equation}
|
||||
%
|
||||
Here, every single component refers to a probabilistic sensor model.
|
||||
The barometer information is evaluated using $p(\vec{o}_t \mid \vec{q}_t)_\text{baro}$,
|
||||
whereby absolute position information is given by $p(\vec{o}_t \mid \vec{q}_t)_\text{ib}$ for
|
||||
\docIBeacon{}s and by $p(\vec{o}_t \mid \vec{q}_t)_\text{wifi}$ for \docWIFI{}.
|
||||
|
||||
%It is well known that finding analytic solutions for densities is very difficult and only possible in rare cases.
|
||||
%Therefore, numerical solutions like Gaussian filters or the broad class of Monte Carlo methods are deployed \cite{sarkka2013bayesian}.
|
||||
Since we assume indoor localisation to be a time-sequential, non-linear and non-Gaussian process,
|
||||
a particle filter is chosen as approximation of the posterior distribution.
|
||||
\commentByFrank{smoothing?}
|
||||
%Within this work the state transition $p(\mStateVec_{t} \mid \mStateVec_{t-1}, \mObsVec_{t-1})$ is used as proposal distribution,
|
||||
%also known as CONDENSATION algorithm \cite{Isard98:CCD}.
|
||||
and therefore similar to \cite{Ebner-16}.
|
||||
Here, we assume a statistical independence of all sensors and every single component refers to a probabilistic sensor model.
|
||||
The barometer information is evaluated using $p(\vec{o}_t \mid \vec{q}_t)_\text{baro}$, whereby absolute position information is given by $p(\vec{o}_t \mid \vec{q}_t)_\text{ib}$ for \docIBeacon{}s and by $p(\vec{o}_t \mid \vec{q}_t)_\text{wifi}$ for \docWIFI{}.
|
||||
|
||||
|
||||
|
||||
|
||||
|
||||
|
||||
Reference in New Issue
Block a user