A practical bias estimation algorithm for multisensor-multitarget tracking

Abstract

Bias estimation or sensor registration is an essential step in ensuring the accuracy of global tracks in multisensor-multitarget tracking. Most previously proposed algorithms for bias estimation rely on local measurements in centralized systems or tracks in distributed systems, along with additional information like covariances, filter gains or targets of opportunity. In addition, it is generally assumed that such data are made available to the fusion center at every sampling time. In practical distributed multisensor tracking systems, where each platform sends local tracks to the fusion center, only state estimates and, perhaps, their covariances are sent to the fusion center at non-consecutive sampling instants or scans. That is, not all the information required for exact bias estimation at the fusion center is available in practical distributed tracking systems. In this paper, a new algorithm that is capable of accurately estimating the biases even in the absence of filter gain information from local platforms is proposed for distributed tracking systems with intermittent track transmission. Through the calculation of the Posterior Cramér-Rao lower bound and various simulation results, it is shown that the performance of the new algorithm, which uses the tracklet idea and does not require track transmission at every sampling time or exchange of filter gains, can approach the performance of the exact bias estimation algorithm that requires local filter gains.

Optimal estimation of sensor biases for asynchronous multi-sensor data fusion

“…To overcome this difficulty, researchers have exploited a priori knowledge of the target motion model such as the nearly-constant-velocity motion model [18]. Based on this side information, recursive Bayesian approaches for jointly estimating target states and sensor biases have been proposed in [20,21,26,32,34,35]. In particular, estimating sensor biases from asynchronous measurements based on an approximated linear measurement model was studied in [20,21,34,35].…”

“…Based on this side information, recursive Bayesian approaches for jointly estimating target states and sensor biases have been proposed in [20,21,26,32,34,35]. In particular, estimating sensor biases from asynchronous measurements based on an approximated linear measurement model was studied in [20,21,34,35]. However, the approximation procedure in these approaches requires the sensor biases to be small.…”

Multisensor-multitarget bearing-only sensor registration

“…BIAS MODELING IN CARTESIAN COORDINATES As discussed in Section II, bearing-only biases are in the form of the additive constants. Although additive constant biases have already been dealt with in the case of radar measurements [42] and [43], it is not possible to generate similar pseudomeasurements with bearing-only data directly. One way to formulate pseudo-measurements with biases is to model in Cartesian coordinates.…”

“…Besides, the presence of sensor biases that are often unaccounted for can degrade the estimation results significantly. Most of the works on bias estimation have been about radar tracking (see [20], [42], [43] and the references therein) or using other measurements besides bearing information [6], [7]. For example, when the elevation information is available, one can estimate the offset biases as in [7], [14].…”

On Universal Sensor Registration

“…Sensitivity of the estimators in the above contributions was addressed in [5]. An approach of bias calibration using partial Kalman filter data was recently proposed in [6]. A comprehensive literature survey summarizing most of the Kalman filter-based methods may be found in [7].…”