Research On Gaussian Process Regression Based Extended Target Tracking

With the development of high-resolution sensor technology,people are no longer satisfied with obtaining the kinematic state of targets,but also expect to recognize the types of targets,which requires to estimate the shapes of targets.For the extended target tracking algorithms,the target shape can be accurately estimated only by using a suitable measurement model.On the other hand,as for multiple extended target tracking in the presence of clutter,sensors usually receive measurements from both targets and clutter at the same time.A large number of targets and measurements may easily lead to combinatorial explosion,which affects the real-time performance of the filter and even leads to divergence.On the basis of Gaussian process single extended target measurement model,a Gaussian process covariance function that is specially designed for tracking axisymmetric target is proposed in this thesis.Aiming at multiple extended target tracking in clutter environment,a novel data association method is proposed to realize the accurate and fast tracking of multiple extended target.The main works are as follows:1.Research on Gaussian process regression based single extended target tracking algorithm.On the basis of nonlinear filtering,the extended target measurement model and the corresponding measurement equation based on Gaussian process regression are studied.The radial function is used to describe the target extent.Then,by combining the nonlinear filtering algorithm,single extended target tracking in ideal environment is realized.2.Research on axisymmetric extended target tracking method.According to the features of Gaussian process,the influence of covariance function on Gaussian process regression is researched.In practical applications,most aerial targets are axisymmetric.Therefore,aiming at this kind of target,several assumptions about radial functions of targets are proposed in this thesis,which lead to a novel Gaussian process covariance function.In addition,the improvement of the proposed covariance function over the existing ones is analyzed.Simulation results indicate that the use of proposed covariance function can significantly improve the estimation accuracy of target position and extent.3.Research on multiple extended target tracking algorithm in the presence of clutter and missed detection.The thesis studies and analyzes the problem that traditional data association algorithm has too much computation when dealing with multiple extended target tracking.On the basis of linear-time marginal association probabilities calculation,the truncated joint probabilistic data association(TJPDA)algorithm is proposed.It eliminates a large number of association events that do not need to be considered,which effectively reduces the runtime of algorithm.The experimental results demonstrate that in multiple extended target tracking,the GP-TJPDA is more accurate than linear-time JPDA algorithm in target extent estimation and runs faster than the RHM-PHD filter.By combining the proposed covariance function in this thesis,the GP-TJPDA algorithm can track multiple aerial extended targets fast and accurately.