An adaptive virtual measurement STGP algorithm for self-occluded 3D extended object tracking with time-varying shapes

This paper proposes an adaptive virtual measurement Spatio-Temporal Gaussian Process (STGP) algorithm to track a self-occluded 3D object with time-varying shape. Firstly, the STGP is extended to 3D time-varying shape modeling using point cloud data. The temporal correlation is induced to model the shape evolution through equivalent state-space representation of temporal covariance function. Secondly, in order to mitigate self-occlusion effects, an adaptive virtual measurement model is developed in which virtual measurements are generated based on centroid symmetry approximation. The virtual measurement covariance is constructed from sensor noise via symmetric transformation and error propagation, and is further adaptively adjusted using a correction factor. The boundedness of the correction factor is rigorously proven by the stochastic Lyapunov stability analysis. The effectiveness of the proposed method is evaluated in simulation.