Track association and fusion with heterogeneous local trackers

Optimization and Cooperative …, 2009

In this work, we propose a new data processing architecture as well as track association and fusion algorithms to improve target classification and tracking accuracy using distributed and, possibly, legacysensor platforms. We present a robust data fusion algorithm that can incorporate target classes/types at the fusion center when receiving sensor reports and/or local tracks. We aim to tackle the following technical challenges in feature aided tracking.

—In air defense systems, multiple radars with overlapping area of coverage are deployed for surveillance and tracking of aircraft. In such a situation, multiple tracks on the same target may appear. The problem of determining whether two tracks have originated from the same target is termed as Track-to-Track Association (T2TA). This process is followed by a Track-to-Track Fusion (T2TF) process which is done to fuse the tracks of multiple radars. Many multiple radar systems employ centralized T2TA and T2TF which takes place in a fusion center while the radars and their trackers are remotely located from the fusion center. Ideally, the radar trackers must supply target state estimates as well as the covariance and cross-covariance to the fusion center for optimal T2TA and T2TF. However, in case of legacy radars, only the target state estimates are provided to the fusion center. We propose a suboptimal but simple-to-implement technique based on weighted averages for T2TA and T2TF of legacy tracks. In the proposed technique, weighted averages are based on the reliability of the radar, track quality and deviation of target trajectory estimate from its smoothed value.

2008 11th International Conference on Information Fusion, 2008

Feature aided tracking can often yield improved tracking performance over the standard radar tracking with positional measurements alone. However, the complexity of the tracker may dramatically increase due to the inclusion of the target feature state. In this paper, we study the situation where the target feature is a constant or slowly varying parameter with respect to the target state and can be observed together with the target position. We consider using such target feature data for data association which is a significant problem and dominates the outcomes of multi-target tracking in clutter. Extra target discrimination is obtained by computing a joint measurement likelihood which is typically used in a PDA framework. This idea is demonstrated via an example where the target down-range extent measurement is incorporated into a standard IPDA tracker to resolve closely spaced targets in clutter. A simple target extent model is therefore proposed. Our results indicate that when us...

2003

In this paper, we consider a general twosensor, track-to-track association problem, in which an unknown number of targets is modeled as an independent, identically distributed (i.i.d.) system of random elements in a given target state space, while the a priori probability distribution of the total number of targets is not necessarily Poisson. We will show that, in order to accommodate not-necessarily-Poisson distributions, we need to modify the well-known, commonly-used track-to-track association hypothesis evaluation formula, by adding an extra multiplier that is a function of the hypothesized number of the detected targets. In order for this multiplier to be constant, thereby allowing us to use the commonly used track association metric, the Poisson assumption is not only sufficient but also necessary. A general multiple target tracking problem is a dynamic state estimation (filtering) problem in which the system state is that of a set of an unknown number of objects and the obser...

Target tracking problems are theoretically interesting, because the origins of the measurements are not identified. Data association is one of the key techniques on tracking with radar. The problem of data association for target tracking in a cluttered environment with linear target model and non-linear measurement model will be discussed. Firstly, evidences are constructed based on spherical coordinates. Then, the association decisions are constructed according to nearest neighbor and probabilistic data association methods. The simulation results show that the latter method has better performance than the former. Moreover, the results will be compared to linear target tracking, which is really common in data association techniques and it will be shown that there will be a slight decrease in performance of target tracking with nonlinear measurement model.

2006

In this paper we present a benchmark problem for data association based on a real-world networked surveillance system, and compare the behavior of several multidimensional assignment (MDA) algorithms. The problem consists of a set of N s observers which transmit track/event reports to a fusion center through a particular (real-world based) communication network among one of N n networks. The network discards the observer's track identity (ID), replacing it by a network-generated ID and the observer ID, thus losing the information on the origin of the tracks sent by each observer. The solution approach developed in this paper consists of a hierarchical decomposition of the problem. This hierarchical approach first eliminates the redundancy introduced by the communication network by using an MDA algorithm per each observer present, and then using another MDA algorithm to choose which 'non-redundant' reports to fuse. This decomposition drastically reduces the dimensionality of the problem from N s £ N n to N s problems of dimension N n and one of dimension N s .

IEEE Transactions on Signal Processing, 2021

Proceedings of the 37th IEEE Conference on Decision and Control (Cat. No.98CH36171), 1998

Abstract A tactical pilot typically experiences difficulty in maintaining accurate identification on multiple-interacting targets in the presence of clutter. We propose a multilevel feature-based association (MFBA) algorithm to aid a pilot in a dynamic multi-target environment. We ...

2006 9th International Conference on Information Fusion, 2006

Track fusion over a network of sensors requires association of the tracks before the state estimates can be combined. Track association generally involves two steps: evaluating an association metric to score each track-to-track association hypothesis, and selecting the best assignment between two sets of tracks. In many applications feature-aided track association can provide better performance than association with only kinematic data (e.g., position and velocity) when the target density is high. This paper develops a general association metric to support feature-aided track association that considers similarity in both the feature and kinematic domains. The association metric is based upon the maximum a posteriori probability (MAP) approach and can be used for general target and sensor models. Special forms of the association metric are given for some common situations. Numerical results illustrate the performance of different feature association metrics.

IEEE Transactions on Automatic Control, 2000

Abstract| F or the problem of tracking multiple targets the Joint Probabilistic Data Association (JPDA) approach h a s shown to be very e ective in handling clutter and missed detections. The JPDA, however, tends to coalesce neighbouring tracks and ignores the coupling between those tracks. Fitzgerald has shown that hypothesis pruning may be an e ective way t o prevent track coalescence. Unfortunately, this process leads to an undesired sensitivity to clutter and missed detections, and it does not support any coupling. To improve this situation, the paper follows a novel approach t o combine the advantages of JPDA, coupling and hypothesis pruning into new algorithms. First, the problem of multiple target tracking is embedded into one ltering for a linear descriptor system with stochastic coe cients. Next, for this descriptor system the exact Bayesian and new JPDA lters are derived. Finally, through Monte Carlo simulations, it is shown that these new PDA lters are able to handle coupling and are insensitive to track coalescence, to clutter and to missed detections.