Research On 3D Extended Target Tracking Method From Multiple Perspectives

With the increasing resolution,multiple measurements may be generated from a single target,which makes the traditional point-based tracking methods no longer applicable.The extended target tracking method for estimating target motion state and morphology has become a hot and difficult point in the research.However,the existing extended target only considers a two-dimensional plane with fixed view angle,and it cannot estimate the shape in three-dimensional space,which seriously affects the performance of the target tracking system.Therefore,with the help of the theory of three-dimensional reconstruction,this paper presents a three-dimensional extended target tracking method from multiple perspectives,whose main research contents are as follows:Due to the influence of target posture and radar azimuth change,the statistical characteristics of the scattering source are changed and presented with non-uniform characteristics,which makes it difficult for extended target algorithm based on uniform hypothesis to obtain accurate target shape.An extended target tracking method under non-uniform distribution measurement is proposed,which produces uniformly distributed pseudo-measurements through the contour centroid method and acceptance-rejection sampling.Hierarchical Unscented Kalman Filter(HUKF)is designed to estimate the motion state and shape parameters of the target.The simulation results show that the proposed HUKF method has better-extended target tracking performance than the traditional UKF,and can provide more accurate boundary information for target 3D shape reconstruction.To obtain the accurate three-dimensional spatial structure of the target,a multi-view 3D extended target tracking method based on Visual Hull is presented on HUKF contour estimation.The algorithm binarizes the target contour estimation results from multiple perspectives,solves the three-dimensional voxel distribution structure through reprojection mapping,and finally creates the visible shell of the target using Marching Cubes algorithm to obtain the spatial three-dimensional structure and location information of the target.The simulation results show that the proposed three-dimensional extended target tracking algorithm has well-extended shape reconstruction and location estimation performance for targets of different shapes.Due to the limited viewing angle and incomplete measurement information,it is difficult to reconstruct an accurate three-dimensional shape of the target.A distributed voxel fusion three-dimensional extended target tracking method is proposed.Based on the change of viewing angle caused by relative motion,the contour sequences obtained by distributed sensors are reconstructed with voxels to get the local three-dimensional structure of the target,and then the intersection of them is fused to get the initial voxel arrangement.Finally,the complete three-dimensional model of the target is solved by the interpolation of the contour surface.The simulation results show that the algorithm improves the accuracy of three-dimensional modeling further than Visual Hull's multi-view 3D extended target tracking method.