An interacting multiple model correntropy Kalman filter approach for unmanned aerial vehicle localisation

This paper compares a conventional interacting multiple model Kalman filter (IMM-KF) filter and an interacting multiple models with maximum correntropy Kalman filter (IMM-MCKF). A nonlinear UAV dynamics model was used to compare these two methods. The compared filters estimated the position of the UAV under the noise distribution. Although KF has reliable accuracy, MCKF has got better results under non-Gaussian or mixed distributions. At this point, these filters have been investigated under maneuver and non-maneuver motion, and it is known that better advantages will be shown when both filters are used in the IMM. These filters have been compared under non-Gaussian distributions, and the Student’s-T distribution has been selected as a non-Gaussian type. The performance validation and testing stages are carried out with variable degrees of freedom, and scaling matrix factors for the Student’s-T distributions have been used. Results from simulation tests from 3000 independent Monte-Carlo runs are presented. In these experiments, UAV models and UAV trajectory results have been used.