Improvement of Multiple Ground Targets Tracking with GMTI Sensor and Fusion of Identification Attributes

Multiple ground targets (MGT) tracking is a challenging problem in real environment because of partial observations, high traffic density, the maneuverability of targets, the clutter and the low target detection probabilities. Most of current MGT trackers use GMTI (Ground Moving Target Indicator) sensor, since this sensor provides the range- rate measurement (Doppler) aside classical position measurements. This helps the tracking algorithms and the preliminary ground target classification. Advanced algorithms include exogeneous information like road network and terrain topography. In this paper, we develop a new improved VS-IMM (Variable Structure Interacting Multiple Model) algorithm for GMTI tracking which includes the stop-move target maneuvering model, contextual information (on-off road model, road network constraints), and identification information arising from classifiers coupled with the GMTI sensor. The identification information is integrated to the likelihood of each hypothesis of our SB-MHT and allows to solve efficiently most of ambiguities that can arise mainly at road intersections or after a target maneuver when leaving the road, or with undetected ground targets after few scans. We maintain aside each target track a set of ID hypotheses with their committed beliefs which are updated on real time with classifier decisions through target type tracker based on a proportional conflict redistribution fusion rule. The advantage of such a new approach is to deal precisely and efficiently with the identification attribute information available as it comes by taking into account its inherent uncertainty/non-specificity and possible high auto-conflict.