Novel adaptive Kalman filtering and fuzzy track fusion approach for real time applications

The track fusion combines individual tracks formed by different sensors. Tracks are usually obtained by Kalman filter (KF), since it is suitable for real-time application. The KF is an optimal linear estimator when the measurement noise has a Gaussian distribution with known covariance. However, in practice, some of the sensors do not have these properties, and the traditional KF is not an optimal estimator. In this paper, a novel adaptive Kalman filter (NAKF) is proposed. In this approach, the measurement noise covariance is adjusted by using an introduced simple mathematical function of one variable, called the degree of matching (DoM), where it is defined on the basis of covariance matching technique. In the fusion structure, each measurement coming from each sensor is fed to a NAKF. So n sensors and n NAKFs will work together in parallel. To obtain the fused track, a fuzzy track fusion method is also proposed. In this method, a fuzzy weight is assigned to each track based on the values of DoM, and another variable is generated by using the track quality function. The fuzzy weight of each track shows the degree of confidence of each track among others. Finally, defuzzification using the center of gravity can obtain the fused track. The NAKF and the proposed fusion methods have very simple structures with low computational cost and accurate performance. Hence, they are suitable to be used in real-time applications. Simulation results show not only the effectiveness and accuracy of using the NAKF in track estimation, but also the good performance of the proposed track fusion method in compare with the other common fusion methods such as simple convex combination and Bar-Shalom/Campo state vector combination methods.