Attention-Based Maneuver-Aware Tracking Network for Maneuvering Target Tracking

Maneuvering target tracking is always a challenging problem due to the complexity of target motion state. Especially with highly maneuvering target, existing tracking algorithms struggle to swiftly and accurately respond to the sudden changes in target motion. In this letter, it is proposed for the first time that we should pay attention to the non-stationarity of maneuvering trajectory segments, and a self-attention-based maneuvering target tracking framework is developed. Specially, the proposed method first resorts to the maneuvering factors acquired from the mean and variance of trajectory segments to extract non-stationary maneuvering information, and then relies on the long-term dependence between observation segments to achieve maneuvering target tracking. Additionally, to enhance the robustness of the tracking model under the sudden change of motion patterns, a local feature embedding module is proposed to extract dynamically the local motion information of maneuvering target. Numerical experiments demonstrate the superiority of our proposed method over advanced deep learning-based maneuvering target tracking methods in enhancing modeling capabilities and improving robustness under abrupt changes in motion patterns.