<p>The advancement of modern high-resolution radars enables the resolution of multiple measurements originating from various unidentified sources on an object. As a result, the object must be treated as an extended object possessing a spatial extent. The Gaussian processes (GPs) extended object model characterizes the distribution of unknown functions through mean and kernel functions, effectively learning the radial function that describes the object’s shape and capturing this extended form comprehensively. Consequently, it has attracted considerable interest in both military and civilian applications. However, the tracking of maneuvering extended objects in dynamic and complex scenarios presents substantial challenges. To address this, an approach for maneuvering extended object tracking based on improved Gaussian processes with adaptive contouring is proposed to enhance the robustness of the tracking system. Initially, Strong Tracking Filter (STF) theory is employed to improve system robustness. Nevertheless, the STF encounters limitations in addressing the decline in estimation accuracy of the object’s orientation angle due to the complex relationship between motion and shape. Therefore, a modification technique for the object’s orientation angle based on motion direction is introduced. This method utilizes the motion direction to correct the interactive output orientation angle, reducing orientation angle estimation errors. Furthermore, the object’s shape may experience sudden changes due to shifts in posture or mission requirements. Unfortunately, traditional GPs parameters remain static, often resulting in tracking failures when initial parameters fail to align with the current scenarios. To mitigate this, an adaptive contouring method based on shape context is proposed, which adjusts system parameters adaptively according to changes in estimation result convergence. The efficacy of the proposed algorithm is demonstrated through simulation experiments conducted in diverse scenarios.</p>