Gaussian Mixtures Proposal Density in Particle Filter for Track-Before-Detect

Size: px
Start display at page:

Download "Gaussian Mixtures Proposal Density in Particle Filter for Track-Before-Detect"

Transcription

1 12th International Conference on Information Fusion Seattle, WA, USA, July 6-9, 29 Gaussian Mixtures Proposal Density in Particle Filter for Trac-Before-Detect Ondřej Straa, Miroslav Šimandl and Jindřich Duní Department of Cybernetics, Research Centre Data-Algorithms-Decision Maing, Faculty of Applied Sciences, University of West Bohemia, Czech Republic Abstract The paper deals with state estimation for the trac-before-detect approach using the particle filter. The focus is aimed at the trac initiation proposal density of the particle filter which considerably affects estimate quality. The goal of the paper is to design a proposal based on a Gaussian mixture using a ban of extended Kalman filters. This leads to root mean square error lower than that achieved by usual simple trac initiation proposals. Due to application of several developed techniques reducing computational requirements of the designed proposal, the Gaussian mixture particle filter also achieves lower computational requirements than ordinary particle filter. Performance of the proposed Gaussian mixture trac initiation proposal in the particle filter is demonstrated in a numerical example. Keywords: Tracing, nonlinear filtering, estimation, tracbefore-detect, Gaussian mixture, proposal density, particle filtering 1 Introduction Classical tracing approaches estimating target state consider target measurements, typically position, range, bearing and so forth, that are extracted by thresholding from the output of the sensor signal processing unit. The purpose of the thresholding is to simplify subsequent target state estimation by eeping only the data exceeding the threshold and thus reducing the data flow. These approaches are not suitable for tracing targets with low signal-to-noise ratio (SNR), typically stealthy military aircraft and cruise missiles, for which thresholding has an undesirable effect of disregarding potentially useful data [1]. To cope with tracing low SNR targets, the tracing approach woring with raw unthresholded data is used. This approach thus has to simultaneously detect and trac target and is nown in literature as the trac-beforedetect (TBD) approach [2]. Due to nonlinearity and/or non-gaussianity of the model considered in the TBD approach, it is necessary to use a global nonlinear estimation technique which computes probability density function (pdf) of the target state. The particle filter () [3] is a global state estimation method providing an approximate of the pdf of the state estimate in the form of a set of particles and corresponding weights. The has dominated in recursive nonlinear state estimation due to its easy implementation in very general settings and cheap and formidable computational power. This is also the reason for popularity within the TBD approach [4], [1], [5]. This paper focuses on the proposal density of the for the TBD approach. The proposal density, which serves for particles sampling, considerably affects quality of estimates and therefore its careful design may increase quality of the estimates. Traditional designs of the proposals use no or little information (bootstrap proposal) coming from the available measurement in the process of sampling the particles. Utilizing full information from the measurement in the proposal density should improve the positioning of the particles and consequently estimate quality. A natural choice for a better proposal is to use the composite approach [6], which is based on utilizing another nonlinear estimation technique to obtain a proposal density. The paper follows the aforementioned idea and designs a Gaussian mixture (GM) proposal density for trac initiation (TI). The proposal is computed by a ban of extended Kalman filters (EKF s). The proposal density approximates the true filtering pdf which implies a very simple and fast computation of the weights corresponding to the particles. The paper also discusses reduction of high computational requirements of the proposed algorithm. The reduction of the requirements follows two directions. The former consists in limiting the number of measurements processed by each EKF and the latter in limiting the number of EKF s utilized and precomputing a large amount of data processed in the EKF s. By utilizing these techniques, in many cases the proposed Gaussian mixture () achieves lower computation demands than the with bootstrap proposal density [7]. The paper is organized as follows: Section 2 introduces the model considered in the TBD approach, Section 3 provides an algorithm of the for the TBD and Section 4 presents the GM proposal for the. In Section 5 the reduction of computational requirements of the GM proposal is discussed, Section 6 provides a numerical example demonstrating the properties of the proposed in compari ISIF 27

2 son with the with special focus on choice of parameters, root mean square (RMS) error and computational aspects and Section 7 concludes the paper together with several remars on performance of the. 2 Trac-Before-Detect The state of the target at time instant is defined by position (x, y ) and velocity (ẋ, ẏ ) of the target in the x and y directions and by the return intensity I of the target, which is considered to be unnown. Thus the target state is given by x = [x, ẋ, y, ẏ, I ] T. The target state evolves according to the discrete-time linear Gaussian model [1] as x +1 = Fx + e, (1) where e is a zero mean white Gaussian noise with covariance matrix Q, i.e. p(e ) = N {e :, Q}. As the constant velocity process model is considered, the transition matrix F is given by F = 1 T 1 1 T 1 1 and the covariance matrix of the state noise by q s T 3 /3 q s T 2 /2 q s T 2 /2 q s T Q = q s T 3 /3 q s T 2 /2 q s T 2 /2 q s T, q i T (3) where T is the sampling period, q s is the power spectral density of the acceleration noise in the spatial dimensions and q i is the power spectral density of the noise in the rate of change of target return intensity. The prior pdf of the target state upon its appearance at time is denoted as p b (x ). Indication of the target presence in the measured data is modelled through a target existence variable E [8], with E = 1 if the target is present in data or E = if the target is absent. Probability of the target presence is modelled by a Marov chain with two states and the following transition matrix [ ] 1 Pb P = b, (4) Pd 1 P d where P b is probability of target birth and P d is probability of target death. The measurements obtained as a result of sensor signal processing are in the form of a sequence of images [1]- Ch. 11. The measurement z at time instant is assumed to be a two-dimensional image consisting of n x cells in the (2) x direction and n y cells in the y direction. Hence, each measurement z is a set {z } n x,n y i=1, j=1 of values of measured intensity at each cell. Each cell contains a contribution of the target denoted as h (x ) and noise denoted as v. If the target is not present, only contribution of the noise is considered, i.e. z = { h (x ) + v, if target present v, if target not present. The measurement noise v is considered to be a white zero-mean Gaussian noise with variance R for all cells, i.e. p(v ) = N {v :, R}. Note that the noise is independent between cells. The nonlinear function h (x ) representing contribution of the target to each cell is given by h (x ) = x y I 2π 2 exp ( ( xi x ) 2 + ( y j y ) ) (5), (6) where x and y represent the size of a cell in the x and y directions respectively. It can be seen that the spread of the measurement contributed by the target is modelled by a two-dimensional Gaussian distribution. As the state estimation methods assume the measurement to be a vector, let us denote the measurement z and the set of measurement functions {h (x )} n x,n y i=1, j=1 in (6) staced up as columns of n x n y elements as z and h(x ), respectively. The aim of estimation here is to find an estimate of the state x based on all the measurements up to time in the form of the conditional pdf p(x z ), where z = [z T, zt 1,...,zT ]T. This pdf completely describes the filtering estimate of x, conditioned by z. For the purpose of estimation, the state of the target x must be extended by the existence variable E representing the presence of the target, which is also unnown. The extended state x is defined as x = [x T, E ] T. Thus the state estimation problem of x changes to the hybrid estimation problem of x with continuous part x and discrete part E. The general solution to the state estimation problem is given by the Bayesian recursive relations which provide filtering estimate of the state x in the form of the filtering probability density function p( x z ) as p(x, E =1 z ) = p(z x, E =1)p(x, E =1 z 1 ) p(z z 1. ) (7) 271

3 The predictive pdf p(x, E =1 z 1 ) in (7) is given as p(x,e =1 z 1 ) = p(x, E =1 x 1, E 1 =1) p(x 1, E 1 =1 z 1 )dx 1 + p(x, E =1, E 1 = z 1 ) = p(x x 1, E =1, E 1 =1)[1 P d ] p(x 1, E 1 =1 z 1 )dx 1 + p b (x )P b. (8) The transition pdf p(x x 1, E = 1, E 1 = 1) in (8) is given by the state equation (1) as p(x x 1, E = 1, E 1 = 1) = N {x : Fx 1, Q} (9) and the measurement pdf p(z x, E =1) in (7) is given by the measurement equation (5) in the following form n x n y p(z x, E = 1) = N {z : h (x ), R} (1) i=1 j=1 p(z x, E = ) = p(z E = ) = n x n y = N {z :, R}. (11) i=1 j=1 Note that for the purpose of state x estimation by means of the it is suitable approximate the measurement pdf p(z x, E = 1) to reduce the computational requirements. As the target affects only the cells within the vicinity of its location (x, y ), it is possible to approximate the measurement pdf as p(z x, E = 1) i C(x ) j C(y ) i / C(x ) j / C(y ) N {z : h (x ), R} N {z :, R}, (12) where C(x ) and C(y ) represent sets of cell indices within the vicinity of the target position in the x direction and y direction, respectively. 3 Particle filter for trac-beforedetect The idea of the in nonlinear state estimation is to approximate the target filtering pdf p( x z ) by the empirical filtering pdf r N ( x z ), which is given by N random samples of the state { x (i) }N i=1 and associated weights {w ( x (i) )}N i=1, w ( ), N i=1 w ( x (i) ) = 1, as r N ( x z ) = N w ( x (i) )δ( x x (i) ), (13) i=1 where δ( ) is the Dirac function defined as δ(x) = for x = and δ(x)dx = 1. The general algorithm of the can be found in [9]. In the case of the TBD approach the particles are divided at each time instant into two groups, the alive particles, with E = 1, and dead particles, with E =. The state part x of the dead particles is not defined. For alive particles the state part x is drawn from the proposal densities π(x x (i) 1, E(i) 1 = 1, z ) for living particles (i.e. E (i) = 1 and E (i) 1 = 1) and π b(x E (i) 1 =, z ) for newborn particles (i.e. E (i) = 1 and E (i) 1 = ). Consider an empirical pdf r N ( x 1 z 1 ) with uniform weights {w 1 ( x (i) 1 )}N i=1 and N a out of N total samples being alive. For the purpose of the TBD approach, the fundamental steps of the general algorithm are modified as (see [1]-Ch.11): Sampling: Each of N a alive particles will die with probability P d and survive with probability 1 P d, i.e. The target existence variable of N a P d randomly chosen alive particles that will die will become zero, E (i) =. The remaining N a (1 P d ) alive particle will survive with E (i) = 1 and their state part x (i) is drawn from the proposal density π, i.e. x (i) π(x x (i) 1, E(i) 1 =1, z ) Each of (N N a ) dead particles will remain dead with probability 1 P b and be born with probability P b, i.e. The target existence variable of (N N a )(1 P b ) randomly chosen dead particles that will remain dead will be zero, E (i) =. The remaining (N N a )P b dead particle will be born with E (i) = 1 and the state x (i) will be drawn from the TI proposal density, i.e. x (i) π b (x E (i) 1 =, z ) Note that the existence variable part E (i) the corresponding state part x (i) x (i). together with form the new particle Weighting: The weights of the particles that survived or were born, i.e. with E (i) = 1, are computed as Survived particles w ( x (i) ) = p(z x (i), E(i) = 1)p(x (i) x(i) 1, E =1, E 1 =1) π(x (i) x(i) 1, E(i) 1 =1, z ) (14) 272

4 Newborn particles w ( x (i) ) = p(z x (i), E(i) = 1)p b (x (i) ) π b (x (i) E(i) 1 =, z ) (15) The weights of the particles that remained dead or have died are given by w ( x (i) ) = p(z E (i) = ). (16) All the weights are then normalized, i.e. w ( x (i) ) = w ( x (i) )/ N j=1 w ( x ( j) ), Resampling: Generate a new set { x (i) }N i=1 by resampling with replacement N times from { x (i) }N i=1 with probability P( x (i) = x(i) ) = w ( x (i) ) and set w ( x (i) ) = N 1. Replace the sets { x (i) }N i=1 and {w ( x (i) )}N i=1 by the resampled sets { x (i) }N i=1 and {w ( x (i) )}N i=1 respectively. To greatly simplify the relations for weight computation (14, 15 and 16), which can be demanding especially for large number of measurement cells, it is possible to utilize the approximation (12) of p(z x (i) = 1) as follows: All, E(i) the measurement pdf s in (14, 15 and 16) are divided by p(z E (i) = ) (11), i.e. they are replaced by the lielihood ratio l(z x (i), E(i) ) = p(z x (i), E(i) )/p(z E (i) = ). This implies the following simplifications. For survived and newborn particles (see [1]): l(z x (i), E(i) = 1) j) N {z(l, : h (l, j) (x (i) ), R} l C(x ) j C(y ) l C(x ) j C(y ) l C(x ) j C(y ) exp ( For dead particles: N {z(l, j) :, R} h(l, j) (x (i) )[h(l, j) (x (i) 2R = ) 2z(l, j) ] ). (17) l(z x (i), E(i) = ) = 1. (18) The estimate of probability of target existence is then given by Ni=1 E (i) ˆP = (19) N and the target state estimate by Ni=1 E (i) ˆx = x(i) Ni=1 E (i). (2) Choices of the proposal densities π and π b greatly influence performance of the. A survey of proposal densities can be found in [6]. The next subsection will discuss choice of proposal densities π(x x (i) 1, E(i) 1 = 1, z ) and π b (x (i) E(i) 1 =, z ) for the TBD approach. Choice of the proposal density In the TBD approach there are two proposal densities, π(x x (i) 1, E(i) 1 = 1, z ) for the surviving particles further denoted as transition proposal and π b (x (i) E(i) 1 =, z ) for the newborn particles, further denoted as TI proposal. As far as the transition proposal is concerned, the simplest choice of the proposal is the transition pdf (9) which does not involve the last available measurement z. In [5] the Rao-Blacwellisation (RB) technique was used to increase efficiency of the proposal density for slightly different sensor model. Further, to reduce the computational demands of the RB technique, which are high in this case as for each particle the proposal density must be calculated, gating techniques were utilized. In [7] it was argued that although the measurement can be used to improve the proposal, due to expected low target SNR s there is little to gain. As far as the TI proposal is concerned, a comparison of three proposal densities was presented in [7]. The first proposal distributes the particle uniformly in the state space, i.e. the TI proposal is equal to the prior TI pdf p b (x ). The second proposal uses available measurement so that particle positions are distributed uniformly within the highest intensity cells. The remaining particle components, i.e. velocities in both directions and intensity are sampled from the prior TI pdf. The third proposal represents a combination of the first two proposals, i.e. half of the particles are distributed uniformly using the prior TI pdf and the other half are places in the highest intensity cells. This paper deals with the TI proposal density only and the goal is to propose a sophisticated TI proposal which achieves high estimation quality with computational demands comparable to the three above mentioned proposals. As was mentioned in Introduction, to use more information from the measurement in the proposal, the composite approach has been chosen. The approach is based on choosing another nonlinear filtering technique and use it as a generator of the proposal density. Based on the comparison published in [6] and the nature of the TBD approach, the GM proposal [1] based on the GM method [11] has been chosen. 4 Gaussian mixture proposal density To design the GM proposal, it is necessary to specify the prior TI pdf p b (x ) in terms of the GM. The state part of each particle consists of position, velocity and intensity components which are considered to be independent for trac initiation. Therefore the prior TI pdf p b (x ) can be written as p b (x ) = p b (x, y )p b (ẋ, ẏ )p b (I ). As far as the velocity components are concerned, in the stage of trac initiation they do not influence the measurement and therefore will be drawn from the prior TI pdf p b (ẋ, ẏ ) and 273

5 omitted in the GM proposal. The other components, i.e. position and intensity, directly influence the measurement and therefore will be present in the GM proposal. The prior TI pdf p b (x, y ) for position components is considered to be uniform within the whole area covered by the measurement, i.e. [, x n x ] in the x-direction and [, y n y ] in the y-direction. The GM approximation p b (x, y ) of this prior TI pdf is given by a rectangular and equally spaced grid of N x N y points, which are distributed within the whole rectangular area covered by the measurement, representing predictive means and by corresponding diagonal predictive covariance matrices which are equal for all grid points. The uniform pdf p b (x, y ) may be approximated by p b (x, y ) up to arbitrary accuracy. The grid may either cover the whole area or, based on the idea used in [7], cover only the highest intensity cells. The choice of will be discussed in Section 5. As far as the intensity component I is concerned, its prior TI distribution is also uniform within the interval [I min, I max ] and therefore for the purpose of the GM proposal it will be approximated by a grid of N I equally spaced points representing predictive means and by corresponding variances, which are equal for all intensity grid points. The complete grid of predictive means is constructed by the Cartesian product of the position grid points and intensity grid points. The weights of all terms of the GM approximation are equal and the predictive pdf p(x, y, I ) is given as p(x, y, I ) = N N x N y N I i=1 x y I 1 N x N y N I ˆx i, : ŷ i,, Î i, P x, P y, P I,, (21) where [ˆx i, ŷ i, ]T is a position grid point with predictive covariance matrix x, [ P ] P y,, and Î i, is an intensity grid point with predictive variance P I,. For notational purposes the predictive covariance matrix in (21) will be denoted as P Ṫo compute the GM proposal density in the form N x N y N I p(x, y, I z ) = N x y I : N x N y N I ˆx i, ŷ i, Î i, i=1 α i,, P i,, (22) EKF s are used to compute the filtering means [ ˆx i,, ŷ i,, Î i, ] T and covariance matrices Pi, as ˆx i, ˆx i, ˆx ŷ i, = ŷ i, i, + K i, (z h ŷ Î i, Î i, i, ) (23) Î i, P i, =(I K i, H i, )P, (24) where K i, is the Kalman gain given by K i, = P HT i, (R n x n y + H i, P HT i, ) 1, (25) with R nx n y being n x n y n x n y dimensional diagonal matrix with variance R on its diagonal and H i, being the Jacobian of h ( ) at [ˆx i,, ŷ i,, Î i, ]T. The weights α i, are computed as ˆx i, α i, = N z : h ( ŷ i, ), R nx n y + H P Î i, i, HT. (26) and normalized so that N x N y N I i=1 α i, = 1. It is clear that running N x N y N I extended Kalman filters for n x n y dimensional measurements, even if they may be run in parallel, requires a large amount of computational resources. The following section discusses possibilities of reduction of the requirements. The GM (22) is used as the proposal for the newborn particles in the sampling stage of the. As the proposal (22) corresponds to the filtering pdf, the weights of the samples drawn from (22) have to evaluate only the discrepancy between the uniform prior p b (x ) and the predictive pdf (21), which is negligible for sufficiently large N x, N y and N I. This fact saves computational effort of the weighting step within the. 5 Computational efficiency There are several causes of high computational requirements of the. First, the number of measurements processed by each EKF is n x n y and consequently there is the same number of rows in the Jacobian H. Second, the number of predictive means in (21) is quite high when considering the whole position grid. Remedies to both causes of high computational requirements are discussed in the following subsections. 5.1 Reducing number of measurements processed As has been already mentioned, the target, if present, influences only a few cells within its vicinity. This means that the Jacobian H is sparse and so is the Kalman gain (25). Therefore the influence of the cells that are not in the vicinity of the target is negligible. Note that this fact was confirmed by experiments. The indices (l, j) of the cells at the vicinity of each predictive mean at (x i,, y i, ) position can 274

6 be specified as l C(x) = { x i, s,..., x i, + s}, j C(y) = { y i, s,..., y i, + s} where is the ceiling function, s is the user-specified parameter implying that each EKF now uses (2s + 1) 2 measurements instead of n x n y with (2s + 1) 2 n x n y. If the predictive mean lies close to the borders of monitored area, i.e. closer than s cells in arbitrary direction, the number of the measurements utilized by the corresponding EKF is even lower than (2s + 1) 2. This poses a problem when computing the weight α i, using (26) as its value strongly depends on the actual number of measurements used. To eliminate this dependency, the modified relation for weight α i, computation is used: α i, = Nz l C(x) j C(y) N : h (l, j) ( j) z(l, ˆx i, ŷ i, Î i, ), P z,i,, (27) where P z,i, = R nx n y + H i, P HT i, and N z [(s + 1) 2, (2s + 1) 2 ] is the actual number of measurement used for computation of this weight. 5.2 Reducing number of Kalman filters and precomputation When specifying the number of predictive GM terms in (21), it must be noted that use of too many terms is as performance-wise bad as that of too few terms [2] due to excessive competition from the unnecessary GM terms. In this case it is possible to reduce the number of position grid points using the idea borrowed from [7] based on choosing a specified number of the highest intensity cells. Similarly, in the case of GM proposal it is possible to consider only the position grid points located within the cells with highest intensity. A discussion of the choice of is a part of the numerical example. Another way of reducing the computational requirements of the GM proposal is to precompute as much terms as possible. The Kalman gain (25), filtering covariance matrix P i,, and the term R nx n y + H i, P HT i, are among the terms that can be computed in advance as they are independent of the measurement z. The filtering mean [ˆx i,, ŷ i,, Î i, ] T and the weight α i, must be calculated at each time step using (23) and (27), respectively. 6 Numerical example To assess performance of the and compare it with the with bootstrap TI proposal density based on distributing particles uniformly within the highest intensity cells, further denoted as the, both filters processed the same measurements. The comparison is based on 1 Monte Carlo experiments. As the and differ only in the TI proposal, the comparison is based on performance at the TI stage, i.e. when the target appears in the scene. For both the and N = 1 particles were used. The parameters of the measurement equation (5) defined as sizes of the cells are x = y = 1, the deviation of the target spread function =.7, the variance of the measurement noise R = 3. Each frame of data consisted of n x = n y = 64 cells in the x and y directions, the intensity I was considered within range [1, 3] which corresponds to the pea target strengths between approximately.69 and 1.23dB in SNR. Note that the whole range of intensity considered belongs to the TBD with low SNR. This is illustrated by the fact that even for the highest intensity considered (I = 3) due to the measurement noise in average 71 measurement cells have higher intensity than the cell containing the target. The true position of the target is drawn from the prior TI pdf p b (x, y ), i.e. uniformly within the whole region. As the velocity components of the state do not influence the measurement and are generated from their prior TI pdf for both and, they will not be considered in the experiments. The considered 4 position grid points within each measurement cell that was taen into account with their variance P x, = P y, =.5 and 6 intensity grid points with their variance equal to P I, = 4. They constitute the predictive GM. Experiment I This experiment deals with performance of the and for different SNR s. The performance was measured in terms of root mean square error (RMSE) of the position components and intensity component RM SE(x) = 1 1 (ˆx (m) x (m)) 1 2 (28) m=1 RM SE(I) = 1 1 (Î (m) I (m)) 1 2, (29) m=1 where ˆx (m) and Î (m) are estimates provided by the or according to (2) at the mth simulation. Note that error of the position in the y direction was not evaluated as it was equal to the error in the x direction. Both the and considered = 1 cells of the highest measurement intensities. The results are depicted in Fig. 1. As far as the position estimate is considered, the clearly outperforms the by approximately 1% for all SNR s. The intensity estimate of the achieves smaller error than that of the for larger SNR. For small SNR the intensity estimate is better. The experiments indicated that a significant increase of intensity estimate quality for low SNR can be achieved by pruning the insignificant terms from the GM proposal but this issue requires further analysis. 275

7 RMSE(I) = 1 RMSE(I) SNR = dB RMSE(x) SNR [db] SNR [db] Figure 1: RMSE of and for varying SNR Experiment II This experiment deals with performance of the and for different choices of. The results are depicted in Fig. 2 for 1.52dB SNR (I = 11), in Fig. 3 for 6.71dB (I = 2)and in Fig. 4 for 9.94dB SNR (I = 29). The shaded area highlights a usually considered range of. As far as quality of the intensity estimates of both and is considered, it can be seen that for low SNR (1.52dB) with increasing number of measurements ( ) the RMSE decreases for both and. According to the figure, the intensity estimate quality of is worse than that of the in this case but if the is very low (up to = 1), the outperforms the. For medium (6.71dB) and high (9,94dB) SNR s increasing brings decrease of the RMSE only for the first few measurements. Further increase of causes on contrary increase of the RMSE. This behavior of the intensity estimates is in accord with [2]. For high SNR only few of the highest intensity cell measurements brings useful information for both the and. The remaining measurements are unnecessary and pose excessive competition for the terms corresponding to the quality measurements. For low SNR each measurement cell brings only little information and no competition exists in this case. As far as position estimates are considered, for approximately < 2 RMSE of both and decrease with increasing with the achieving smaller error than the, whereas for approximately > 2 the RMSE of both and increases with increasing. This fact is also a manifestation of too many predictive terms. Computational requirements To compare computational requirements of the and, computational time was calculated for varying number of measurement cells with the highest intensity. The RMSE(x) Figure 2: RMSE of the and for varying and 1.52 db SNR RMSE(I) RMSE(x) SNR = 6.717dB Figure 3: RMSE of the and for varying and 6.71 db SNR experiments were executed within the MATLAB environment on a 2GHz PC. The results consisting in computational time of sampling 1 samples from the proposal densities of the (blue solid line), (red dashed line) and time spent within the for actual drawing samples and weighting, i.e. without calculating the GM terms by (22), (27), (green dotted line) are depicted on Fig. 5. The results show that for < 2 the achieves even lower computational complexity than the. For > 2 the is computationally more demanding than the due to increasing number of terms of the filtering pdf (22) that are computed. It must be pointed out that in this case the effect of the precomputation and the reduction of the number of EKF s proposed in Section 5 is reduction of computational demands by more than 87%. Also it should be noted that for the considering < 1 is usually sufficient for achieving high estimate quality. 276

8 RMSE(I) RMSE(x) SNR = 9.938dB Figure 4: RMSE of the and for varying and 9.94 db SNR time [s] SNR = 9.938dB sampling and weighting part of Figure 5: Computational time of the and for varying and 9.94dB SNR 7 Conclusion The paper focused on TI proposal density of the in the TBD approach to estimation of the target state. The aim was to utilize more information from the available measurement to increase quality of the particles drawn from the TI proposal in comparison with the usual bootstrap TI proposal. To compute the new proposal, the GM method based on a ban of EKF s was used. The quality of the target position estimate in terms of the RMSE achieved by the is better than that of the. Also quality of the target intensity estimate achieved by the is better than that of the (with the exception of SNR lower than 6dB which can be prevented by substantial pruning of the GM proposal). The computational efficiency of the was increased by choosing only small number of measurements with high intensity, limiting number of GM terms and precomputing as many terms of the EKF s as possible. The achieved results showed that with a careful choice of parameters the achieves higher quality of estimates with lower computational complexity than the. The future wor will be devoted to analysis of possibilities of GM pruning techniques which, according to experiments carried out, achieves better intensity estimates for very low SNR. 8 Acnowledgments The wor was supported by the Ministry of Education, Youth and Sports of the Czech Republic, project No. 1M572, by the Czech Science Foundation, project GA12/8/442 and by the University of West Bohemia, project POSTDOC-9. References [1] B. Ristic, S. Arulampalam, and N. Gordon, Beyond the Kalman Filter: Particle Filters for Tracing Applications. Artech House, 24. [2] X. R. Li and Y. Bar-Shalom, Multiple-model estimation with variable structure, IEEE Transactions on Automatic Control, vol. 41, no. 4, pp , [3] A. Doucet, N. De Freitas, and N. Gordon, Sequential Monte Carlo Methods in Practice. Springer, 21, ch. An Introduction to Sequential Monte Carlo Methods, (Ed. Doucet A., de Freitas N., and Gordon N.). [4] Y. Boers and J. Driessen, Particle filter based detection for tracing, in Proceedings of the American Control Conference, 21, vol. 6, 21. [5] A. Ooi, A. Doucet, B. N. Vo, and B. Ristic, Particle filter for tracing linear Gaussian target with nonlinear observations, in Proc. SPIE, vol. 596, 23. [6] O. Straa and M. Simandl, Sampling densities of particle filter: a survey and comparison, in Proceedings of the 26th American Control Conference (ACC). New Yor: AACC, 27, pp [7] M. G. Rutten, B. Ristic, and N. Gordon, A comparison of particle filters for recursive trac-before-detect, in 7th Internatinal Conference on Information Fusion (FUSION), 25, pp [8] D. Musici, R. Evans, and S. Stanovic, Integrated probabilistic data association, IEEE Trans. on Automatic Control, vol. 39, no. 6, pp , [9] A. Doucet, N. De Freitas, and N. Gordon, Eds., Sequential Monte Carlo Methods in Practice. Springer, 21, (Ed. Doucet A., de Freitas N., and Gordon N.). [1] J. H. Kotecha and P. M. Djuric, Gaussian sum particle filtering, in IEEE Transactions on Signal Processing, ser. 51, no. 1, 23. [11] K. Ito and K. Xiong, Gaussian filters for nonlinear filtering problems, IEEE Trans. on Automatic Control, vol. 45, no. 5, pp ,

Sensor Fusion: Particle Filter

Sensor Fusion: Particle Filter Sensor Fusion: Particle Filter By: Gordana Stojceska stojcesk@in.tum.de Outline Motivation Applications Fundamentals Tracking People Advantages and disadvantages Summary June 05 JASS '05, St.Petersburg,

More information

SIGMA POINT GAUSSIAN SUM FILTER DESIGN USING SQUARE ROOT UNSCENTED FILTERS

SIGMA POINT GAUSSIAN SUM FILTER DESIGN USING SQUARE ROOT UNSCENTED FILTERS SIGMA POINT GAUSSIAN SUM FILTER DESIGN USING SQUARE ROOT UNSCENTED FILTERS Miroslav Šimandl, Jindřich Duní Department of Cybernetics and Research Centre: Data - Algorithms - Decision University of West

More information

Performance Evaluation of Local State Estimation Methods in Bearings-only Tracking Problems

Performance Evaluation of Local State Estimation Methods in Bearings-only Tracking Problems 14th International Conference on Information Fusion Chicago, Illinois, USA, July -8, 211 Performance Evaluation of Local State Estimation ethods in Bearings-only Tracing Problems Ondřej Straa, Jindřich

More information

BAYESIAN MULTI-TARGET TRACKING WITH SUPERPOSITIONAL MEASUREMENTS USING LABELED RANDOM FINITE SETS. Francesco Papi and Du Yong Kim

BAYESIAN MULTI-TARGET TRACKING WITH SUPERPOSITIONAL MEASUREMENTS USING LABELED RANDOM FINITE SETS. Francesco Papi and Du Yong Kim 3rd European Signal Processing Conference EUSIPCO BAYESIAN MULTI-TARGET TRACKING WITH SUPERPOSITIONAL MEASUREMENTS USING LABELED RANDOM FINITE SETS Francesco Papi and Du Yong Kim Department of Electrical

More information

Analytically-Guided-Sampling Particle Filter Applied to Range-only Target Tracking

Analytically-Guided-Sampling Particle Filter Applied to Range-only Target Tracking Analytically-Guided-Sampling Particle Filter Applied to Range-only Target Tracing Guoquan P. Huang and Stergios I. Roumeliotis Abstract Particle filtering (PF) is a popular nonlinear estimation technique

More information

Particle Filters. Outline

Particle Filters. Outline Particle Filters M. Sami Fadali Professor of EE University of Nevada Outline Monte Carlo integration. Particle filter. Importance sampling. Degeneracy Resampling Example. 1 2 Monte Carlo Integration Numerical

More information

Multitarget Particle filter addressing Ambiguous Radar data in TBD

Multitarget Particle filter addressing Ambiguous Radar data in TBD Multitarget Particle filter addressing Ambiguous Radar data in TBD Mélanie Bocquel, Hans Driessen Arun Bagchi Thales Nederland BV - SR TBU Radar Engineering, University of Twente - Department of Applied

More information

RAO-BLACKWELLISED PARTICLE FILTERS: EXAMPLES OF APPLICATIONS

RAO-BLACKWELLISED PARTICLE FILTERS: EXAMPLES OF APPLICATIONS RAO-BLACKWELLISED PARTICLE FILTERS: EXAMPLES OF APPLICATIONS Frédéric Mustière e-mail: mustiere@site.uottawa.ca Miodrag Bolić e-mail: mbolic@site.uottawa.ca Martin Bouchard e-mail: bouchard@site.uottawa.ca

More information

A Tree Search Approach to Target Tracking in Clutter

A Tree Search Approach to Target Tracking in Clutter 12th International Conference on Information Fusion Seattle, WA, USA, July 6-9, 2009 A Tree Search Approach to Target Tracking in Clutter Jill K. Nelson and Hossein Roufarshbaf Department of Electrical

More information

A FEASIBILITY STUDY OF PARTICLE FILTERS FOR MOBILE STATION RECEIVERS. Michael Lunglmayr, Martin Krueger, Mario Huemer

A FEASIBILITY STUDY OF PARTICLE FILTERS FOR MOBILE STATION RECEIVERS. Michael Lunglmayr, Martin Krueger, Mario Huemer A FEASIBILITY STUDY OF PARTICLE FILTERS FOR MOBILE STATION RECEIVERS Michael Lunglmayr, Martin Krueger, Mario Huemer Michael Lunglmayr and Martin Krueger are with Infineon Technologies AG, Munich email:

More information

Rao-Blackwellized Particle Filter for Multiple Target Tracking

Rao-Blackwellized Particle Filter for Multiple Target Tracking Rao-Blackwellized Particle Filter for Multiple Target Tracking Simo Särkkä, Aki Vehtari, Jouko Lampinen Helsinki University of Technology, Finland Abstract In this article we propose a new Rao-Blackwellized

More information

RAO-BLACKWELLIZED PARTICLE FILTER FOR MARKOV MODULATED NONLINEARDYNAMIC SYSTEMS

RAO-BLACKWELLIZED PARTICLE FILTER FOR MARKOV MODULATED NONLINEARDYNAMIC SYSTEMS RAO-BLACKWELLIZED PARTICLE FILTER FOR MARKOV MODULATED NONLINEARDYNAMIC SYSTEMS Saiat Saha and Gustaf Hendeby Linöping University Post Print N.B.: When citing this wor, cite the original article. 2014

More information

in a Rao-Blackwellised Unscented Kalman Filter

in a Rao-Blackwellised Unscented Kalman Filter A Rao-Blacwellised Unscented Kalman Filter Mar Briers QinetiQ Ltd. Malvern Technology Centre Malvern, UK. m.briers@signal.qinetiq.com Simon R. Masell QinetiQ Ltd. Malvern Technology Centre Malvern, UK.

More information

Incorporating Track Uncertainty into the OSPA Metric

Incorporating Track Uncertainty into the OSPA Metric 14th International Conference on Information Fusion Chicago, Illinois, USA, July 5-8, 211 Incorporating Trac Uncertainty into the OSPA Metric Sharad Nagappa School of EPS Heriot Watt University Edinburgh,

More information

Fisher Information Matrix-based Nonlinear System Conversion for State Estimation

Fisher Information Matrix-based Nonlinear System Conversion for State Estimation Fisher Information Matrix-based Nonlinear System Conversion for State Estimation Ming Lei Christophe Baehr and Pierre Del Moral Abstract In practical target tracing a number of improved measurement conversion

More information

Lecture 7: Optimal Smoothing

Lecture 7: Optimal Smoothing Department of Biomedical Engineering and Computational Science Aalto University March 17, 2011 Contents 1 What is Optimal Smoothing? 2 Bayesian Optimal Smoothing Equations 3 Rauch-Tung-Striebel Smoother

More information

A Sequential Monte Carlo Approach for Extended Object Tracking in the Presence of Clutter

A Sequential Monte Carlo Approach for Extended Object Tracking in the Presence of Clutter A Sequential Monte Carlo Approach for Extended Object Tracing in the Presence of Clutter Niolay Petrov 1, Lyudmila Mihaylova 1, Amadou Gning 1 and Dona Angelova 2 1 Lancaster University, School of Computing

More information

The Mixed Labeling Problem in Multi Target Particle Filtering

The Mixed Labeling Problem in Multi Target Particle Filtering The Mixed Labeling Problem in Multi Target Particle Filtering Yvo Boers Thales Nederland B.V. Haasbergerstraat 49, 7554 PA Hengelo, The Netherlands yvo.boers@nl.thalesgroup.com Hans Driessen Thales Nederland

More information

The Unscented Particle Filter

The Unscented Particle Filter The Unscented Particle Filter Rudolph van der Merwe (OGI) Nando de Freitas (UC Bereley) Arnaud Doucet (Cambridge University) Eric Wan (OGI) Outline Optimal Estimation & Filtering Optimal Recursive Bayesian

More information

Sigma Point Belief Propagation

Sigma Point Belief Propagation Copyright 2014 IEEE IEEE Signal Processing Letters, vol. 21, no. 2, Feb. 2014, pp. 145 149 1 Sigma Point Belief Propagation Florian Meyer, Student Member, IEEE, Ondrej Hlina, Member, IEEE, and Franz Hlawatsch,

More information

State Estimation by IMM Filter in the Presence of Structural Uncertainty 1

State Estimation by IMM Filter in the Presence of Structural Uncertainty 1 Recent Advances in Signal Processing and Communications Edited by Nios Mastorais World Scientific and Engineering Society (WSES) Press Greece 999 pp.8-88. State Estimation by IMM Filter in the Presence

More information

NON-LINEAR NOISE ADAPTIVE KALMAN FILTERING VIA VARIATIONAL BAYES

NON-LINEAR NOISE ADAPTIVE KALMAN FILTERING VIA VARIATIONAL BAYES 2013 IEEE INTERNATIONAL WORKSHOP ON MACHINE LEARNING FOR SIGNAL PROCESSING NON-LINEAR NOISE ADAPTIVE KALMAN FILTERING VIA VARIATIONAL BAYES Simo Särä Aalto University, 02150 Espoo, Finland Jouni Hartiainen

More information

Suppression of impulse noise in Track-Before-Detect Algorithms

Suppression of impulse noise in Track-Before-Detect Algorithms Computer Applications in Electrical Engineering Suppression of impulse noise in Track-Before-Detect Algorithms Przemysław Mazurek West-Pomeranian University of Technology 71-126 Szczecin, ul. 26. Kwietnia

More information

Estimating the Shape of Targets with a PHD Filter

Estimating the Shape of Targets with a PHD Filter Estimating the Shape of Targets with a PHD Filter Christian Lundquist, Karl Granström, Umut Orguner Department of Electrical Engineering Linöping University 583 33 Linöping, Sweden Email: {lundquist, arl,

More information

The Kernel-SME Filter with False and Missing Measurements

The Kernel-SME Filter with False and Missing Measurements The Kernel-SME Filter with False and Missing Measurements Marcus Baum, Shishan Yang Institute of Computer Science University of Göttingen, Germany Email: marcusbaum, shishanyang@csuni-goettingende Uwe

More information

Conditional Posterior Cramér-Rao Lower Bounds for Nonlinear Sequential Bayesian Estimation

Conditional Posterior Cramér-Rao Lower Bounds for Nonlinear Sequential Bayesian Estimation 1 Conditional Posterior Cramér-Rao Lower Bounds for Nonlinear Sequential Bayesian Estimation Long Zuo, Ruixin Niu, and Pramod K. Varshney Abstract Posterior Cramér-Rao lower bounds (PCRLBs) 1] for sequential

More information

Data Assimilation for Dispersion Models

Data Assimilation for Dispersion Models Data Assimilation for Dispersion Models K. V. Umamaheswara Reddy Dept. of Mechanical and Aerospace Engg. State University of New Yor at Buffalo Buffalo, NY, U.S.A. venatar@buffalo.edu Yang Cheng Dept.

More information

Density Approximation Based on Dirac Mixtures with Regard to Nonlinear Estimation and Filtering

Density Approximation Based on Dirac Mixtures with Regard to Nonlinear Estimation and Filtering Density Approximation Based on Dirac Mixtures with Regard to Nonlinear Estimation and Filtering Oliver C. Schrempf, Dietrich Brunn, Uwe D. Hanebeck Intelligent Sensor-Actuator-Systems Laboratory Institute

More information

Randomized Unscented Kalman Filter in Target Tracking

Randomized Unscented Kalman Filter in Target Tracking Randomized Unscented Kalman Filter in Target Tracking Ondřej Straka, Jindřich Duník and Miroslav Šimandl Department of Cybernetics, Faculty of Applied Sciences, University of West Bohemia, Univerzitní

More information

F denotes cumulative density. denotes probability density function; (.)

F denotes cumulative density. denotes probability density function; (.) BAYESIAN ANALYSIS: FOREWORDS Notation. System means the real thing and a model is an assumed mathematical form for the system.. he probability model class M contains the set of the all admissible models

More information

Cardinality Balanced Multi-Target Multi-Bernoulli Filtering Using Adaptive Birth Distributions

Cardinality Balanced Multi-Target Multi-Bernoulli Filtering Using Adaptive Birth Distributions Cardinality Balanced Multi-Target Multi-Bernoulli Filtering Using Adaptive Birth Distributions Stephan Reuter, Daniel Meissner, Benjamin Wiling, and Klaus Dietmayer Institute of Measurement, Control, and

More information

A New Nonlinear Filtering Method for Ballistic Target Tracking

A New Nonlinear Filtering Method for Ballistic Target Tracking th International Conference on Information Fusion Seattle, WA, USA, July 6-9, 9 A New Nonlinear Filtering Method for Ballistic arget racing Chunling Wu Institute of Electronic & Information Engineering

More information

Combined Particle and Smooth Variable Structure Filtering for Nonlinear Estimation Problems

Combined Particle and Smooth Variable Structure Filtering for Nonlinear Estimation Problems 14th International Conference on Information Fusion Chicago, Illinois, USA, July 5-8, 2011 Combined Particle and Smooth Variable Structure Filtering for Nonlinear Estimation Problems S. Andrew Gadsden

More information

A NEW FORMULATION OF IPDAF FOR TRACKING IN CLUTTER

A NEW FORMULATION OF IPDAF FOR TRACKING IN CLUTTER A NEW FRMULATIN F IPDAF FR TRACKING IN CLUTTER Jean Dezert NERA, 29 Av. Division Leclerc 92320 Châtillon, France fax:+33146734167 dezert@onera.fr Ning Li, X. Rong Li University of New rleans New rleans,

More information

A Comparison of Particle Filters for Personal Positioning

A Comparison of Particle Filters for Personal Positioning VI Hotine-Marussi Symposium of Theoretical and Computational Geodesy May 9-June 6. A Comparison of Particle Filters for Personal Positioning D. Petrovich and R. Piché Institute of Mathematics Tampere University

More information

Random Finite Set Methods. for Multitarget Tracking

Random Finite Set Methods. for Multitarget Tracking Random Finite Set Methods for Multitarget Tracing RANDOM FINITE SET METHODS FOR MULTITARGET TRACKING BY DARCY DUNNE a thesis submitted to the department of electrical & computer engineering and the school

More information

Content.

Content. Content Fundamentals of Bayesian Techniques (E. Sucar) Bayesian Filters (O. Aycard) Definition & interests Implementations Hidden Markov models Discrete Bayesian Filters or Markov localization Kalman filters

More information

Probability Hypothesis Density Filter for Multitarget Multisensor Tracking

Probability Hypothesis Density Filter for Multitarget Multisensor Tracking Probability Hypothesis Density Filter for Multitarget Multisensor Tracing O. Erdinc, P. Willett, Y. Bar-Shalom ECE Department University of Connecticut ozgur, willett @engr.uconn.edu ybs@ee.uconn.edu Abstract

More information

Particle Filter Track Before Detect Algorithms

Particle Filter Track Before Detect Algorithms Particle Filter Track Before Detect Algorithms Theory and Applications Y. Boers and J.N. Driessen JRS-PE-FAA THALES NEDERLAND Hengelo The Netherlands Email: {yvo.boers,hans.driessen}@nl.thalesgroup.com

More information

Multi-Target Particle Filtering for the Probability Hypothesis Density

Multi-Target Particle Filtering for the Probability Hypothesis Density Appears in the 6 th International Conference on Information Fusion, pp 8 86, Cairns, Australia. Multi-Target Particle Filtering for the Probability Hypothesis Density Hedvig Sidenbladh Department of Data

More information

Acceptance probability of IP-MCMC-PF: revisited

Acceptance probability of IP-MCMC-PF: revisited Acceptance probability of IP-MCMC-PF: revisited Fernando J. Iglesias García, Mélanie Bocquel, Pranab K. Mandal, and Hans Driessen. Sensors Development System Engineering, Thales Nederland B.V. Hengelo,

More information

Target tracking and classification for missile using interacting multiple model (IMM)

Target tracking and classification for missile using interacting multiple model (IMM) Target tracking and classification for missile using interacting multiple model (IMM Kyungwoo Yoo and Joohwan Chun KAIST School of Electrical Engineering Yuseong-gu, Daejeon, Republic of Korea Email: babooovv@kaist.ac.kr

More information

NONLINEAR STATISTICAL SIGNAL PROCESSING: A PARTI- CLE FILTERING APPROACH

NONLINEAR STATISTICAL SIGNAL PROCESSING: A PARTI- CLE FILTERING APPROACH NONLINEAR STATISTICAL SIGNAL PROCESSING: A PARTI- CLE FILTERING APPROACH J. V. Candy (tsoftware@aol.com) University of California, Lawrence Livermore National Lab. & Santa Barbara Livermore CA 94551 USA

More information

AN EFFICIENT TWO-STAGE SAMPLING METHOD IN PARTICLE FILTER. Qi Cheng and Pascal Bondon. CNRS UMR 8506, Université Paris XI, France.

AN EFFICIENT TWO-STAGE SAMPLING METHOD IN PARTICLE FILTER. Qi Cheng and Pascal Bondon. CNRS UMR 8506, Université Paris XI, France. AN EFFICIENT TWO-STAGE SAMPLING METHOD IN PARTICLE FILTER Qi Cheng and Pascal Bondon CNRS UMR 8506, Université Paris XI, France. August 27, 2011 Abstract We present a modified bootstrap filter to draw

More information

A Gaussian Mixture PHD Filter for Nonlinear Jump Markov Models

A Gaussian Mixture PHD Filter for Nonlinear Jump Markov Models A Gaussian Mixture PHD Filter for Nonlinear Jump Marov Models Ba-Ngu Vo Ahmed Pasha Hoang Duong Tuan Department of Electrical and Electronic Engineering The University of Melbourne Parville VIC 35 Australia

More information

AUTOMOTIVE ENVIRONMENT SENSORS

AUTOMOTIVE ENVIRONMENT SENSORS AUTOMOTIVE ENVIRONMENT SENSORS Lecture 5. Localization BME KÖZLEKEDÉSMÉRNÖKI ÉS JÁRMŰMÉRNÖKI KAR 32708-2/2017/INTFIN SZÁMÚ EMMI ÁLTAL TÁMOGATOTT TANANYAG Related concepts Concepts related to vehicles moving

More information

Extension of the Sparse Grid Quadrature Filter

Extension of the Sparse Grid Quadrature Filter Extension of the Sparse Grid Quadrature Filter Yang Cheng Mississippi State University Mississippi State, MS 39762 Email: cheng@ae.msstate.edu Yang Tian Harbin Institute of Technology Harbin, Heilongjiang

More information

The Cardinality Balanced Multi-Target Multi-Bernoulli Filter and its Implementations

The Cardinality Balanced Multi-Target Multi-Bernoulli Filter and its Implementations PREPRINT: IEEE TRANSACTIONS ON SIGNAL PROCESSING, VOL. 57, NO. 2, PP. 49-423, 29 1 The Cardinality Balanced Multi-Target Multi-Bernoulli Filter and its Implementations Ba-Tuong Vo, Ba-Ngu Vo, and Antonio

More information

Tracking and Identification of Multiple targets

Tracking and Identification of Multiple targets Tracking and Identification of Multiple targets Samir Hachour, François Delmotte, Eric Lefèvre, David Mercier Laboratoire de Génie Informatique et d'automatique de l'artois, EA 3926 LGI2A first name.last

More information

EVALUATING SYMMETRIC INFORMATION GAP BETWEEN DYNAMICAL SYSTEMS USING PARTICLE FILTER

EVALUATING SYMMETRIC INFORMATION GAP BETWEEN DYNAMICAL SYSTEMS USING PARTICLE FILTER EVALUATING SYMMETRIC INFORMATION GAP BETWEEN DYNAMICAL SYSTEMS USING PARTICLE FILTER Zhen Zhen 1, Jun Young Lee 2, and Abdus Saboor 3 1 Mingde College, Guizhou University, China zhenz2000@21cn.com 2 Department

More information

Lecture 8: Bayesian Estimation of Parameters in State Space Models

Lecture 8: Bayesian Estimation of Parameters in State Space Models in State Space Models March 30, 2016 Contents 1 Bayesian estimation of parameters in state space models 2 Computational methods for parameter estimation 3 Practical parameter estimation in state space

More information

Gaussian Mixture PHD and CPHD Filtering with Partially Uniform Target Birth

Gaussian Mixture PHD and CPHD Filtering with Partially Uniform Target Birth PREPRINT: 15th INTERNATIONAL CONFERENCE ON INFORMATION FUSION, ULY 1 Gaussian Mixture PHD and CPHD Filtering with Partially Target Birth Michael Beard, Ba-Tuong Vo, Ba-Ngu Vo, Sanjeev Arulampalam Maritime

More information

1 Kalman Filter Introduction

1 Kalman Filter Introduction 1 Kalman Filter Introduction You should first read Chapter 1 of Stochastic models, estimation, and control: Volume 1 by Peter S. Maybec (available here). 1.1 Explanation of Equations (1-3) and (1-4) Equation

More information

Particle Filters. Pieter Abbeel UC Berkeley EECS. Many slides adapted from Thrun, Burgard and Fox, Probabilistic Robotics

Particle Filters. Pieter Abbeel UC Berkeley EECS. Many slides adapted from Thrun, Burgard and Fox, Probabilistic Robotics Particle Filters Pieter Abbeel UC Berkeley EECS Many slides adapted from Thrun, Burgard and Fox, Probabilistic Robotics Motivation For continuous spaces: often no analytical formulas for Bayes filter updates

More information

L09. PARTICLE FILTERING. NA568 Mobile Robotics: Methods & Algorithms

L09. PARTICLE FILTERING. NA568 Mobile Robotics: Methods & Algorithms L09. PARTICLE FILTERING NA568 Mobile Robotics: Methods & Algorithms Particle Filters Different approach to state estimation Instead of parametric description of state (and uncertainty), use a set of state

More information

Nonlinear Measurement Update and Prediction: Prior Density Splitting Mixture Estimator

Nonlinear Measurement Update and Prediction: Prior Density Splitting Mixture Estimator Nonlinear Measurement Update and Prediction: Prior Density Splitting Mixture Estimator Andreas Rauh, Kai Briechle, and Uwe D Hanebec Member, IEEE Abstract In this paper, the Prior Density Splitting Mixture

More information

Multi-Target Tracking in a Two-Tier Hierarchical Architecture

Multi-Target Tracking in a Two-Tier Hierarchical Architecture Multi-Target Tracing in a Two-Tier Hierarchical Architecture Jin Wei Department of Electrical Engineering University of Hawaii at Manoa Honolulu, HI 968, U.S.A. Email: weijin@hawaii.edu Xudong Wang Department

More information

TSRT14: Sensor Fusion Lecture 8

TSRT14: Sensor Fusion Lecture 8 TSRT14: Sensor Fusion Lecture 8 Particle filter theory Marginalized particle filter Gustaf Hendeby gustaf.hendeby@liu.se TSRT14 Lecture 8 Gustaf Hendeby Spring 2018 1 / 25 Le 8: particle filter theory,

More information

Tracking of Extended Objects and Group Targets using Random Matrices A New Approach

Tracking of Extended Objects and Group Targets using Random Matrices A New Approach Tracing of Extended Objects and Group Targets using Random Matrices A New Approach Michael Feldmann FGAN Research Institute for Communication, Information Processing and Ergonomics FKIE D-53343 Wachtberg,

More information

Non-linear and non-gaussian state estimation using log-homotopy based particle flow filters

Non-linear and non-gaussian state estimation using log-homotopy based particle flow filters Non-linear and non-gaussian state estimation using log-homotopy based particle flow filters Muhammad Altamash Khan, Martin Ulme Sensor Data and Information Fusion Department FKIE Fraunhofer, Wachtberg,

More information

Multi-Target Tracking Using A Randomized Hypothesis Generation Technique

Multi-Target Tracking Using A Randomized Hypothesis Generation Technique Multi-Target Tracing Using A Randomized Hypothesis Generation Technique W. Faber and S. Charavorty Department of Aerospace Engineering Texas A&M University arxiv:1603.04096v1 [math.st] 13 Mar 2016 College

More information

Extended Object and Group Tracking with Elliptic Random Hypersurface Models

Extended Object and Group Tracking with Elliptic Random Hypersurface Models Extended Object and Group Tracing with Elliptic Random Hypersurface Models Marcus Baum Benjamin Noac and Uwe D. Hanebec Intelligent Sensor-Actuator-Systems Laboratory ISAS Institute for Anthropomatics

More information

Fast Sequential Monte Carlo PHD Smoothing

Fast Sequential Monte Carlo PHD Smoothing 14th International Conference on Information Fusion Chicago, Illinois, USA, July 5-8, 11 Fast Sequential Monte Carlo PHD Smoothing Sharad Nagappa and Daniel E. Clark School of EPS Heriot Watt University

More information

GMTI Tracking in the Presence of Doppler and Range Ambiguities

GMTI Tracking in the Presence of Doppler and Range Ambiguities 14th International Conference on Information Fusion Chicago, Illinois, USA, July 5-8, 2011 GMTI Tracing in the Presence of Doppler and Range Ambiguities Michael Mertens Dept. Sensor Data and Information

More information

STONY BROOK UNIVERSITY. CEAS Technical Report 829

STONY BROOK UNIVERSITY. CEAS Technical Report 829 1 STONY BROOK UNIVERSITY CEAS Technical Report 829 Variable and Multiple Target Tracking by Particle Filtering and Maximum Likelihood Monte Carlo Method Jaechan Lim January 4, 2006 2 Abstract In most applications

More information

Hierarchical Particle Filter for Bearings-Only Tracking

Hierarchical Particle Filter for Bearings-Only Tracking Hierarchical Particle Filter for Bearings-Only Tracing 1 T. Bréhard and J.-P. Le Cadre IRISA/CNRS Campus de Beaulieu 3542 Rennes Cedex, France e-mail: thomas.brehard@gmail.com, lecadre@irisa.fr Tel : 33-2.99.84.71.69

More information

Particle based probability density fusion with differential Shannon entropy criterion

Particle based probability density fusion with differential Shannon entropy criterion 4th International Conference on Information Fusion Chicago, Illinois, USA, July -8, Particle based probability density fusion with differential Shannon entropy criterion Jiří Ajgl and Miroslav Šimandl

More information

Introduction to Mobile Robotics Bayes Filter Particle Filter and Monte Carlo Localization

Introduction to Mobile Robotics Bayes Filter Particle Filter and Monte Carlo Localization Introduction to Mobile Robotics Bayes Filter Particle Filter and Monte Carlo Localization Wolfram Burgard, Cyrill Stachniss, Maren Bennewitz, Kai Arras 1 Motivation Recall: Discrete filter Discretize the

More information

Nonlinear Estimation Techniques for Impact Point Prediction of Ballistic Targets

Nonlinear Estimation Techniques for Impact Point Prediction of Ballistic Targets Nonlinear Estimation Techniques for Impact Point Prediction of Ballistic Targets J. Clayton Kerce a, George C. Brown a, and David F. Hardiman b a Georgia Tech Research Institute, Georgia Institute of Technology,

More information

Extended Target Tracking with a Cardinalized Probability Hypothesis Density Filter

Extended Target Tracking with a Cardinalized Probability Hypothesis Density Filter Extended Target Tracing with a Cardinalized Probability Hypothesis Density Filter Umut Orguner, Christian Lundquist and Karl Granström Department of Electrical Engineering Linöping University 58 83 Linöping

More information

X. F. Wang, J. F. Chen, Z. G. Shi *, and K. S. Chen Department of Information and Electronic Engineering, Zhejiang University, Hangzhou , China

X. F. Wang, J. F. Chen, Z. G. Shi *, and K. S. Chen Department of Information and Electronic Engineering, Zhejiang University, Hangzhou , China Progress In Electromagnetics Research, Vol. 118, 1 15, 211 FUZZY-CONTROL-BASED PARTICLE FILTER FOR MANEUVERING TARGET TRACKING X. F. Wang, J. F. Chen, Z. G. Shi *, and K. S. Chen Department of Information

More information

Dirac Mixture Density Approximation Based on Minimization of the Weighted Cramér von Mises Distance

Dirac Mixture Density Approximation Based on Minimization of the Weighted Cramér von Mises Distance Dirac Mixture Density Approximation Based on Minimization of the Weighted Cramér von Mises Distance Oliver C Schrempf, Dietrich Brunn, and Uwe D Hanebeck Abstract This paper proposes a systematic procedure

More information

RECURSIVE OUTLIER-ROBUST FILTERING AND SMOOTHING FOR NONLINEAR SYSTEMS USING THE MULTIVARIATE STUDENT-T DISTRIBUTION

RECURSIVE OUTLIER-ROBUST FILTERING AND SMOOTHING FOR NONLINEAR SYSTEMS USING THE MULTIVARIATE STUDENT-T DISTRIBUTION 1 IEEE INTERNATIONAL WORKSHOP ON MACHINE LEARNING FOR SIGNAL PROCESSING, SEPT. 3 6, 1, SANTANDER, SPAIN RECURSIVE OUTLIER-ROBUST FILTERING AND SMOOTHING FOR NONLINEAR SYSTEMS USING THE MULTIVARIATE STUDENT-T

More information

A STATE ESTIMATOR FOR NONLINEAR STOCHASTIC SYSTEMS BASED ON DIRAC MIXTURE APPROXIMATIONS

A STATE ESTIMATOR FOR NONLINEAR STOCHASTIC SYSTEMS BASED ON DIRAC MIXTURE APPROXIMATIONS A STATE ESTIMATOR FOR NONINEAR STOCHASTIC SYSTEMS BASED ON DIRAC MIXTURE APPROXIMATIONS Oliver C. Schrempf, Uwe D. Hanebec Intelligent Sensor-Actuator-Systems aboratory, Universität Karlsruhe (TH), Germany

More information

Two Linear Complexity Particle Filters Capable of Maintaining Target Label Probabilities for Targets in Close Proximity

Two Linear Complexity Particle Filters Capable of Maintaining Target Label Probabilities for Targets in Close Proximity Two Linear Complexity Particle Filters Capable of Maintaining Target Label Probabilities for Targets in Close Proximity Ramona Georgescu and Peter Willett Electrical and Computer Engineering University

More information

An Improved Particle Filter with Applications in Ballistic Target Tracking

An Improved Particle Filter with Applications in Ballistic Target Tracking Sensors & ransducers Vol. 72 Issue 6 June 204 pp. 96-20 Sensors & ransducers 204 by IFSA Publishing S. L. http://www.sensorsportal.co An Iproved Particle Filter with Applications in Ballistic arget racing

More information

A NOVEL OPTIMAL PROBABILITY DENSITY FUNCTION TRACKING FILTER DESIGN 1

A NOVEL OPTIMAL PROBABILITY DENSITY FUNCTION TRACKING FILTER DESIGN 1 A NOVEL OPTIMAL PROBABILITY DENSITY FUNCTION TRACKING FILTER DESIGN 1 Jinglin Zhou Hong Wang, Donghua Zhou Department of Automation, Tsinghua University, Beijing 100084, P. R. China Control Systems Centre,

More information

Gate Volume Estimation for Target Tracking

Gate Volume Estimation for Target Tracking Gate Volume Estimation for Target Tracking Darko Mušicki Mark R. Morelande Dept of Electrical Engineering Dept of Electrical Engineering University of Melbourne University of Melbourne Victoria 30 Victoria

More information

The Kalman Filter ImPr Talk

The Kalman Filter ImPr Talk The Kalman Filter ImPr Talk Ged Ridgway Centre for Medical Image Computing November, 2006 Outline What is the Kalman Filter? State Space Models Kalman Filter Overview Bayesian Updating of Estimates Kalman

More information

Recursive Noise Adaptive Kalman Filtering by Variational Bayesian Approximations

Recursive Noise Adaptive Kalman Filtering by Variational Bayesian Approximations PREPRINT 1 Recursive Noise Adaptive Kalman Filtering by Variational Bayesian Approximations Simo Särä, Member, IEEE and Aapo Nummenmaa Abstract This article considers the application of variational Bayesian

More information

Using the Kalman Filter to Estimate the State of a Maneuvering Aircraft

Using the Kalman Filter to Estimate the State of a Maneuvering Aircraft 1 Using the Kalman Filter to Estimate the State of a Maneuvering Aircraft K. Meier and A. Desai Abstract Using sensors that only measure the bearing angle and range of an aircraft, a Kalman filter is implemented

More information

FUNDAMENTAL FILTERING LIMITATIONS IN LINEAR NON-GAUSSIAN SYSTEMS

FUNDAMENTAL FILTERING LIMITATIONS IN LINEAR NON-GAUSSIAN SYSTEMS FUNDAMENTAL FILTERING LIMITATIONS IN LINEAR NON-GAUSSIAN SYSTEMS Gustaf Hendeby Fredrik Gustafsson Division of Automatic Control Department of Electrical Engineering, Linköpings universitet, SE-58 83 Linköping,

More information

Estimation and Maintenance of Measurement Rates for Multiple Extended Target Tracking

Estimation and Maintenance of Measurement Rates for Multiple Extended Target Tracking Estimation and Maintenance of Measurement Rates for Multiple Extended Target Tracing Karl Granström Division of Automatic Control Department of Electrical Engineering Linöping University, SE-58 83, Linöping,

More information

Distributed estimation in sensor networks

Distributed estimation in sensor networks in sensor networks A. Benavoli Dpt. di Sistemi e Informatica Università di Firenze, Italy. e-mail: benavoli@dsi.unifi.it Outline 1 An introduction to 2 3 An introduction to An introduction to In recent

More information

A Novel Gaussian Sum Filter Method for Accurate Solution to Nonlinear Filtering Problem

A Novel Gaussian Sum Filter Method for Accurate Solution to Nonlinear Filtering Problem A Novel Gaussian Sum Filter Method for Accurate Solution to Nonlinear Filtering Problem Gabriel Terejanu a Puneet Singla b Tarunraj Singh b Peter D. Scott a Graduate Student Assistant Professor Professor

More information

Efficient Monitoring for Planetary Rovers

Efficient Monitoring for Planetary Rovers International Symposium on Artificial Intelligence and Robotics in Space (isairas), May, 2003 Efficient Monitoring for Planetary Rovers Vandi Verma vandi@ri.cmu.edu Geoff Gordon ggordon@cs.cmu.edu Carnegie

More information

9 Multi-Model State Estimation

9 Multi-Model State Estimation Technion Israel Institute of Technology, Department of Electrical Engineering Estimation and Identification in Dynamical Systems (048825) Lecture Notes, Fall 2009, Prof. N. Shimkin 9 Multi-Model State

More information

Human Pose Tracking I: Basics. David Fleet University of Toronto

Human Pose Tracking I: Basics. David Fleet University of Toronto Human Pose Tracking I: Basics David Fleet University of Toronto CIFAR Summer School, 2009 Looking at People Challenges: Complex pose / motion People have many degrees of freedom, comprising an articulated

More information

Extension of the Sliced Gaussian Mixture Filter with Application to Cooperative Passive Target Tracking

Extension of the Sliced Gaussian Mixture Filter with Application to Cooperative Passive Target Tracking 12th International Conference on Information Fusion Seattle, WA, USA, July 6-9, 2009 Extension of the Sliced Gaussian Mixture Filter with Application to Cooperative Passive arget racing Julian Hörst, Felix

More information

Data Assimilation in Variable Dimension Dispersion Models using Particle Filters

Data Assimilation in Variable Dimension Dispersion Models using Particle Filters Data Assimilation in Variable Dimension Dispersion Models using Particle Filters K. V. Umamaheswara Reddy Dept. of MAE venatar@buffalo.edu Yang Cheng Dept. of MAE cheng3@buffalo.edu Tarunraj Singh Dept.

More information

Inferring biological dynamics Iterated filtering (IF)

Inferring biological dynamics Iterated filtering (IF) Inferring biological dynamics 101 3. Iterated filtering (IF) IF originated in 2006 [6]. For plug-and-play likelihood-based inference on POMP models, there are not many alternatives. Directly estimating

More information

Evolution Strategies Based Particle Filters for Fault Detection

Evolution Strategies Based Particle Filters for Fault Detection Evolution Strategies Based Particle Filters for Fault Detection Katsuji Uosaki, Member, IEEE, and Toshiharu Hatanaka, Member, IEEE Abstract Recent massive increase of the computational power has allowed

More information

Tracking an Accelerated Target with a Nonlinear Constant Heading Model

Tracking an Accelerated Target with a Nonlinear Constant Heading Model Tracking an Accelerated Target with a Nonlinear Constant Heading Model Rong Yang, Gee Wah Ng DSO National Laboratories 20 Science Park Drive Singapore 118230 yrong@dsoorgsg ngeewah@dsoorgsg Abstract This

More information

Adaptive Target Tracking in Slowly Changing Clutter

Adaptive Target Tracking in Slowly Changing Clutter Adaptive Target Tracing in Slowly Changing Clutter Thomas Hanselmann, Daro Mušici, Marimuthu Palaniswami Dept. of Electrical and Electronic Eng. The University of Melbourne Parville, VIC 30 Australia {t.hanselmann,

More information

CIS 390 Fall 2016 Robotics: Planning and Perception Final Review Questions

CIS 390 Fall 2016 Robotics: Planning and Perception Final Review Questions CIS 390 Fall 2016 Robotics: Planning and Perception Final Review Questions December 14, 2016 Questions Throughout the following questions we will assume that x t is the state vector at time t, z t is the

More information

Track-before-detect. of this density on the overall performance of the algorithms is investigated. One of the proposals simply places particles

Track-before-detect. of this density on the overall performance of the algorithms is investigated. One of the proposals simply places particles 5 7th International Conference on Information Fusion (FUSION) A Comparison of Particle Filters for Recursive Track-before-detect Mark G. Rutten, Branko Ristic and Neil J. Gordon Intelligence, Surveillance

More information

An introduction to particle filters

An introduction to particle filters An introduction to particle filters Andreas Svensson Department of Information Technology Uppsala University June 10, 2014 June 10, 2014, 1 / 16 Andreas Svensson - An introduction to particle filters Outline

More information

Advanced Computational Methods in Statistics: Lecture 5 Sequential Monte Carlo/Particle Filtering

Advanced Computational Methods in Statistics: Lecture 5 Sequential Monte Carlo/Particle Filtering Advanced Computational Methods in Statistics: Lecture 5 Sequential Monte Carlo/Particle Filtering Axel Gandy Department of Mathematics Imperial College London http://www2.imperial.ac.uk/~agandy London

More information

Estimating Polynomial Structures from Radar Data

Estimating Polynomial Structures from Radar Data Estimating Polynomial Structures from Radar Data Christian Lundquist, Umut Orguner and Fredrik Gustafsson Department of Electrical Engineering Linköping University Linköping, Sweden {lundquist, umut, fredrik}@isy.liu.se

More information

Lecture 2: From Linear Regression to Kalman Filter and Beyond

Lecture 2: From Linear Regression to Kalman Filter and Beyond Lecture 2: From Linear Regression to Kalman Filter and Beyond January 18, 2017 Contents 1 Batch and Recursive Estimation 2 Towards Bayesian Filtering 3 Kalman Filter and Bayesian Filtering and Smoothing

More information