Recursive Joint Track-to-Track Association and Sensor
Nonlinear Bias Estimation Based on Generalized Bayes
Risk
Mengxi Hao
MOE KLINNS Lab
Inst. of Integrated Automation
Xi’an Jiaotong University
Xi’an, China
mx.hao@stu.xjtu.edu.cn
Xianghui Yuan
MOE KLINNS Lab
Inst. of Integrated Automation
Xi’an Jiaotong University
Xi’an, China
xhyuan@mail.xjtu.edu.cn
Chongzhao Han
MOE KLINNS Lab
Inst. of Integrated Automation
Xi’an Jiaotong University
Xi’an, China
czhan@mail.xjtu.edu.cn
Abstract – Track-to-track association and sensor bias
estimation are two important problems in multi-target
multi-sensor tracking system. Track-to-track association
becomes more complex in the presence of sensor bias
and incorrect track association will lead to poor bias
estimation results. Solving these two problems jointly
would be attractive. This paper proposes a recursive joint
track-to-track association and nonlinear bias estimation
algorithm based on the generalized Bayes risk. The
proposed algorithm and the conventional
association-then-estimation algorithm are compared
with the Monte-Carlo simulation. Simulation results
show that the proposed algorithm has better track
association and bias estimation performance than the
conventional algorithm.
Keywords: generalized Bayes risk, recursive,
track-to-track association, sensor bias estimation, joint
decision and estimation.
1 Introduction
Track-to-track association and bias estimation, which
are generally tightly coupled, are of great importance in
multi-target multi-sensor tracking system. The coupling
gives rise to the difficulties.
Much work has been carried out on solving these two
problems separately. The “association-then-estimation”
strategy solves the bias estimation problem and assumes
the association was completely correct. The
“estimation-then-association” strategy solves the
association problem and assumes the bias estimation was
done. These two problems may affect each other and
should be considered jointly.
Several studies have been conducted on joint
track-to-track association and bias estimation problem
(JAE). [1] and [2] proposed a joint MAP bias estimation
and data association algorithm while this algorithm
describes the problem as an nonconvex mixed integer
nonlinear programming problem which is very hard to
solve. A joint association, registration, and fusion
approach based on expectation-maximization (EM) was
proposed in [3]. However it has a drawback that
expectation-maximization algorithm is a batch iterative
algorithm of which the convergence speed is slow when
solving complex cases. [4] proposed an extended product
multi-sensor cardinalized probability hypothesis density
(PM-CPHD) filter for spatial registration and data
association, which leads to a more difficult problem.
The optimal Bayes joint decision and estimation (JDE)
algorithm was proposed in [5] and was used to solve joint
tracking and classification problem in [6]. Moreover, the
optimal Bayes JDE algorithm was improved to recursive
JDE (RJDE) algorithm in [7]. Optimal Bayes JDE
algorithm was applied to solve JAE problem in [8].
However [8] used batch JDE rather than RJDE algorithm,
and only solved linear measurement problem.
In many applications, measurements are obtained
sequentially. So the computational demands of the batch
JDE algorithm will increase with an increase of data. Thus
RJDE algorithm would fit the problem more naturally. In
this paper, we try to apply RJDE algorithm to solve JAE
problem with nonlinear measurement and make it closer to
reality.
This paper is organized as follows. In Section 2 we
briefly describe the sensor bias model and association
problem. Section 3 gives a brief introduction to the RJDE
algorithm. The contribution of this paper is presented in
Section 4, where we use the RJDE method to solve JAE
problem with nonlinear measurement. Section 5 presents
simulation results. Finally, the concluding remarks are
given in Section 6.
2 Problem formulation
In this section, we present the association problem and
the bias estimation problem.
18th International Conference on Information Fusion
Washington, DC - July 6-9, 2015
978-0-9824-4386-6/15/$31.00 ©2015 IEEE 1519