A Gaussian Process Based Extended Target Tracking Method In Polar Coordinates

With the rapid development of high-resolution sensor technology,extended target tracking(ETT)technology has attracted more and more attentions in military and civilian applications,i.e.high-resolution radar detection and automatic driving.ETT not only estimates the kinematic characteristics of the target,but also estimates the contour characteristics.Compared with the traditional point target tracking,ETT has to face more challenges.In high-resolution radar detection applications,targets are usually detected in polar coordinate systems in the presence of strong ground-sea clutter.ETT is a highly complex problem due to sensor noise,missed detections,clutter detections,measurement origin uncertainty,and an unknown and time-varying number of targets.These further increase the difficulty of ETT.In order to address these problems,this thesis considers aerial surveillance problem with high-resolution radar and focuses on the ETT technology using Gaussian process in polar coordinate system.1.In order to address the measurement nonlinearity problem of ETT in polar coordinate system,a Gaussian process based the method with unbiased converted measurement(UCM)is proposed.Firstly,the unbiased converted is used to transform the raw non-linear measurements into linear measurements.This UCM depends on true target state and uses the one-step prediction estimation of the measurement error covariance to eliminate the conversion bias.Then the Gaussian process model is embedded in the probability data association algorithm to estimate the kinematic state and contour state of the extended target.The performance of the proposed method is compared with that the random matrix method.Simulation results show that the proposed method is similar to the random matrix method in the estimation performance of the kinematic state,but can additionally estimate target contours of irregularly shape.2.Calculation of the true measurement error covariance of the UCM requires the true range and bearing,and therefore cannot be done in practice.In order to address the problems,a Gaussian process extended target tracking method based on nonlinear filtering is proposed.Firstly,this method directly establishes the measurement model in polar coordinate system and obtains the corresponding measurement noise covariance.Then,the target contour forgetting factor is combined to establish the target contour transition matrix and the corresponding process noise.Finally,the effective joint tracking gate is established to measurements and obtain the associated probability of related events.Simulations results show that the proposed method has lesser root mean square error(RMSE)and faster rate of convergence.3.In order to track the low signal-to-noise ratio extended target tracking in polar coordinate system,a Gaussian process probability data association method with amplitude information is proposed.Firstly,this method uses the amplitude feature to enhance the target state vector.A detection threshold is used to extract the amplitude information of the extended target from the raw measurement set.Then,the amplitude likelihood ratio is combined with the correlation likelihood ratio to modify the correlation probability.Simulations results show that the proposed method has less center position RMSE and contours RMSE,compared with the Gaussian process probability data association method without amplitude information.