A Gaussian Process Based Non-parametric Model Approach for Maneuvering Target Tracking

In this paper, we consider a non-parametric model approach to solve the maneuvering target tracking problem. Traditional filtering methods rely on accurate parametric models, while most learning based methods require accurate ground-truth training data. To address these challenges, we first establish the process and measurement models for the target using the Gaussian process respectively. Then, online data-based filtering is established where the reduce-rank model formulation of the Gaussian processes and the sliding window mechanism are introduced to reduce the computational complexity. Finally, the Adam optimization algorithm is employed to obtain the estimate of the target state, where the gradient of the likelihood function with respect to the state variable is explicitly derived. The experimental results indicate that in the absence of the parametric models, the maneuvering target tracking performance of the proposed algorithm is close to that of the traditional filtering method with known models, verifying its applicability and robustness for maneuvering target tracking.