3D Extended Target Tracking and Shape Modeling in Clutter

In the environment of clutter and noise, it is difficult to achieve reliable reconstruction of 3D extended target. For extended target, due to the sparsity of the available measurements and the interference from clutter, point cloud data from the extended target cannot be inputted to the Point Completion Network (PCN) directly for shape completion. We combine the Gaussian process and probability data association (GP-PDA) to jointly estimate the kinematics and the shape of the extended target. The proposed GP-PDA filter produces the 3D base points in the contour of the extended target and provides analytical expressions for the representation of the 3D extended target shape. Furthermore, we use the learning-based Point Completion Network to reconstruct the 3D extended target point cloud contour based on the Gaussian base points produced by the GP-PDA filter. The effectiveness of the proposed algorithm is verified by simulation experiments of 3D point cloud data.