Research On Tracking Algorithms For Highly Maneuvering Target

Maneuvering Target tracking has drawn much attention since it was put forward, it is a comprehensive application of sensor technology, computer technology and information processing technology. The major challenge of maneuvering target tracking arises from the target motion-mode uncertainty. There have been many good dynamic models, but none of them is perfect since the diversity of target motion. The standard in state-of-the art tracking algorithm is the Interacting multiple model(IMM) algorithm, which uses a models sets to describe target motion. However, when tracking a highly maneuvering target, the IMM algorithm which uses classical dynamic models and Kalman filter does not work that well.This thesis focuses on methods of tracking highly maneuvering target with jump acceleration, major works include:(1) Aiming at solving the performance decline or divergency problem of a Kalman filter when tracking highly maneuvering targets, an adaptive kalman filter is designed, which is called “Q-switched Kalman”. This filter is able to adjust the process noise covariance matrix according to the maneuver dectection based on the statistical distance of the innovation(or measurement residual). Simulation result shows that the designed filter outperforms the standard Kalman filter when tracking highly maneuvering target.(2) A Dirichlet process based IMM called DP-IMM is proposed. It learns the latent acceleration of a maneuvering target from a set of observation data via Dirichlet process mixture model, then forms a model set, which is used for the IMM algorithm, consists of CV dynamic models with different acceleration as control inputs. Simulations show that the DP-IMM can maintain a good performance when target’s acceleration jumps.