Abstract:

Using the traditional sequential filtering algorithm to track extended targets can lead to an underestimated number of the targets if the multisensor monitoring regions are overlapped. In order to solve this problem, firstly, the measurements are updated by using the current predicted Gaussian components rather than the updated Gaussian components which are obtained through the measurements of any sensor, so that the updated filtering results are independent among sensors. Then the measurements of all sensors are updated. The weights of the Gaussian components whose target positions locate in the overlapped regions are adjusted. Finally, all Gaussian components are pruned and merged. Simulation results show that the improved algorithm is effective and accurate.

Key words: extended target tracking;probability hypothesis density filter;multiple sensors information fusion;state estimation