Target Tracking With Complex Spatial Constraints

Constrained target tracking is important in applications such as ground surveillance and intelligent transportation systems.Target motions are subject to various constraints imposed by external environments,such as roads and channels.The effective extraction,description,and utilization of constraint information can significantly improve target tracking performance.The existing constrained tracking research mainly focus on solving the problem of constrained tracking with completely known and simple trajectory constraints.However,in real applications,such as onroad target tracking,the actual constraint shapes are extremely irregular.The constrained tracking problem may occur,in which only part of the constraint information or no constraint information is available.Therefore,the existing theories and methods are not applicable.In this dissertation,for constrained tracking such as on-road target tracking,modeling of complex spatial constraints and tracking methods are investigated,to solve the problem of how to use spatial constraints to improve tracking performance when complex spatial constraint information is partially known or completely unknown.The main work and contribution are given as follows:1.For the problem where the spatial equality-constrained tracking methods will cause a bias in the state estimation for the situation where the road information is known but the lane is unknown,the modeling of heading constraints and state estimation algorithms are proposed.In this case,the constraint information is extracted based on the phenomenon that a target moves along road direction to improve the target tracking performance.For the basic units describing roads,straight line segments and circular arc segments,the heading constraints are formulated based on the parameter augmentation and state augmentation methods.A discriminating method between the road segments based on the known circular geometric sector information is proposed.The heading constraints are treated as perfect measurements and then are put in the original measurement.The heading-constrained state estimate is realized based on the extended Kalman filter.The posterior Cramer–Rao lower bound of the heading-constrained state estimate is derived.For an on-road maneuvering target,the heading-constrained interacting multiple model tracking methods are developed.Simulation results demonstrate that the proposed headingconstrained target tracking methods can make full use of heading constraint information to improve tracking performance.Compared with the unconstrained algorithm,the root mean square errors of the position and velocity estimates of the proposed method can be reduced by about 51% and 72%,respectively.2.For the problem where the important information that target motions are subjected to constraints imposed by roads cannot be utilized by the conventional tracking methods when no prior information about the regular road is available,a novel tracking algorithm based on separate modeling of target trajectory shape and dynamic characteristics is proposed.In the proposed algorithm,a static model describing the target trajectory is constructed in Cartesian coordinates,while the dynamic models,which have no effect on the trajectory shape,are defined in mileage coordinates.Then,the target trajectory and state are estimated using measurements simultaneously.A sliding window strategy is introduced,where the target trajectory over each window is modeled by first-degree polynomials with parameters to be estimated.In order to determine the target trajectory,an augmented state equation is constructed by augmenting the unknown coefficients of the first-degree polynomials into the base state.The interaction stage of the proposed tracking algorithm starts from the reset parameter vector,which is estimated by least squares.The crosscovariance between the reset parameter vector and the base state vector is derived.The recursive estimation for the parameters of the static model is realized.Simulation results show that the proposed tracking algorithm avoids the problems of model mismatch and information waste that exist in conventional tracking methods when dealing with complex constrained target tracking,and improves the tracking performance.The root mean square errors of the position and velocity estimated by the proposed algorithm can be reduced by more than 37% and 21%,respectively,when compared with the conventional tracking method.The proposed tracking algorithm has relatively more robust performance than those conventional tracking algorithms during the target maneuvers.3.For the problem where a sliding window with linear approximation leads to a mismatched static model when no prior information about road that is meandering and complex is available,a separate modeling based on a sliding window with nonlinear approximation and two novel tracking algorithms are proposed.In the proposed algorithms,the target trajectory is modeled by high-degree polynomials or B-splines.The target dynamic models are defined in mileage coordinates.The trajectory shape and target state are estimated using measurements simultaneously.The function relationship between the initial parameter vector and the Cartesian position can be derived by using the backslash operator.The initial conditions of the augmented states of the different models are calculated by the unscented transformation method.The estimates of the on-road maneuvering target trajectory and motion state are achieved simultaneously based on the least squares and the interacting multiple model in mileage coordinates.By adding a process noise into parametric equations of a static model,the adaptability and robustness of the seconddegree polynomial separation modeling tracking algorithm to irregular trajectory curves are improved.Simulation results demonstrate that the proposed tracking algorithms have higher accuracy and relatively more robust for a maneuvering target than other state-of-the-art algorithms.Compared with the conventional algorithm,the position and velocity estimation errors yielded by the proposed algorithm can be reduced by more than 65% and 60%,respectively.The estimate accuracy of the improved second-degree polynomial separation modeling tracking algorithm with lower computational load is comparable to that of the algorithm with third-degree polynomials.