Modeling and Information Processing
TRACK - Tracking
GMTI (Ground Moving Target Indicator) tracking
Detecting ground targets is a crucial element in military intelligence in battlefield surveillance. Once a target has been detected, the system used can proceed to track it. Tracking can be done using GMTI (Ground Moving Target Indicator) type indicators that can observe all the objects moving in the area of interest. However, the complexity of the problem means that the usual tracking techniques, widely used in the context of aerial targets, cannot be used directly. In practice, contrary to aerial targets, ground targets can maneuver quickly on the road network and in all directions, which may cause the loss of the track. Furthermore, the terrain's topology plays an important role in the accuracy and even the correctness of the information obtained. Vegetation and hill crests can create numerous false alarms; the targets may reappear and disappear due to the terrain masks. Also, road traffic is much denser than that in the airspace, which causes many problems of associations of measurements and false alarms. Therefore, it is evident that knowledge of the terrain's topology is an important factor and this information must be used in the filtering system in order to increase the accuracy of the estimates and reduce the computing time.
Starting from the assumption that all the land moving objects move around the road network, the GMTI measurements supplied by the sensor are assumed to belong to the road network. On the basis of this assumption, several techniques have been studied to take this information into account [1] and [2]. For GMTI tracking, the modeling of a target's dynamics depends on the road on which it is located. The information about the road's orientation gives an indication about the moving object's direction of movement. However, when the target maneuvers on the network, the predetermined dynamic model is no longer suitable. A tracking approach using a filter with multiple models, of the IMM (Interacting Multiple Model) type, can then be considered. The use of the IMM allows us to compensate for the error in estimating the target's state when the target maneuvers on the network (figure 1, video and figure 2). We propose an improvement of the IMM to take terrain topology data into account. This new IMM filter is made up of a variable structure of dynamic models under constraints which is adaptive; it is called VS IMMC.
Figure 1 - Tracing the target's path on the network.
[vidéo]
Figure 2 - The target's dynamic on the network.
Let us consider an IMM with two dynamic models. It is made up of a model 
with low noise for tracking a target that moves at constant speed and a model 
with high noise for tracking a target that maneuvers. From a set 
of road segments (determined for the statistical windowing technique [3]), the system selects the road segment statistically closest to each local estimator of the IMM in order to constrain the dynamic models to the network (i.e. finds the segments 
and 
that can be used at the current time to obtain the models 
and 
. The variation of the set of dynamic models that make up the IMM is therefore linked to the network's topology. Preliminary studies have highlighted the fact that the algorithms behavior is different from that of the usual IMM when this latter is constrained to the network (figure 3). The good performance of the local estimators under constraints allow us to be more demanding about the parameters of the IMM filter (i.e. the transition factor between the models), which leads to an improvement of the VS IMMC filter's performances (figure 4).
Figure 3 - Mean quadratic error in speed and position in 100 simulations with a transition factor between the models of 0.95.
Figure 4 - Mean quadratic error in speed and position in 100 simulations with a transition factor between the models of 0.98.
On the basis of this study, the VS IMMC algorithm will soon be adapted to multi-target tracking techniques [4]. The introduction of contextual information should not only improve the accuracy of tracks but also contribute to reducing the problems of track association.
Publications
[1] B. Pannetier, K. Benameur, V. Nimier, M. Rombaut, Ground Target Tracking With Road Constraint, Proc. SPIE, Sensor Fusion, and Target Recognition XIII, Orlando, FL, August 2004.
[2] B. Pannetier, Techniques pour la trajectographie de cibles terrestres, Rapport Technique Onera, RT 1/09395, janvier 2005.
[3] T. Kirubarajan, Y. Bar-Shalom, K.R. Pattipati, I.Kadar, E. Eadan and B. Abrams, "Tracking Ground Targets with Road Constraints Using an IMM Estimator", Proc. IEEE Aerospace Conf., Snowmass, CO, March 1998.
[4] S. Blackman, R. Popoli, Design and Analysis of Modern Tracking Systems, Artech House, 1999.
