Divers Tracking with Improved Gaussian Mixture Probability Hypothesis Density filter

摘要

The group divers tracking problem with a 2D high-resolution active sonar is studied in this paper. Probability Hypothesis Density (PHD) filter is famous for its good ability in multiple targets tracking. Instead of travelling in a constant velocity motion model, the activity of divers may be, however, affected by the factors such as the destination, activities of surrounded divers and the potential intention of themselves. That is, not only are the motion states of divers correlated with each other but also dependent on the external environment. A solution is proposed to deal with the challenges of a time-varying number of targets, potential interactions by taking advantage of the PHD filter and social forced model (SFM). The diver dynamic model (DDM) is created based on the social force concept. By including the DDM model into the framework of PHD filter, the dependencies from closed group targets and external environments are considered in the recursive Bayesian framework and a different likelihood in prediction stage of a filter can also be obtained. Numerical simulation results show that the proposed method here is able to improve the performance of the PHD filter in the presence of interactions.