A Study On Extended Target Tracking Algorithms In The Environment Of Low SNR

The main function of the radar data processing is to accomplish the target tracking. The traditional target tracking algorithms are based on the assumption that each target can produce at most one measurement at each step, and the whole tracking process is achieved by the track initial, the point-to-track associating and the state filtering. As the radar resolution increases, each target may occupy multiple resolution cells of the sensor. If the traditional algorithms are still used to track the extended targets, it will bring out the problems of high data association complexity and tracking diffusion, so it is very necessary to study the extended target tracking algorithms.This thesis firstly introduced the basic theory of the point target tracking, including the classical track initial algorithms, data association algorithms and filtering algorithms. The track initial algorithms include the logic initial algorithm and the Hough transform algorithm. The data association algorithms include the nearest neighbor data association, the strongest neighbor data association, the probability data association and the joint probability data association. The filtering algorithms include the kalman filtering, the extended kalman filtering, the unscented kalman filtering and the interacting multiple model filtering. The performance of the above algorithms were compared and analysed by the simulation experiments.Then two point target tracking algorithms based on the noise background were discussed. One of them is a sequential bayesian estimation method in the linear non-gaussian system. This method has the advantage of the gaussian sum filtering, and a model order reduction step is introduced to solve the problem of the exponential growth of the system model order number. The other algorithm is named as IMM-TBD, which is based on the track enhancement theory. In this approach,a group of enhancement operators are used to detect the target’s trajectory, and via combining the operators with the interacting multiple model algorithm efficiently, the strong manuvering target tracking problems in the low signal-to-noise ratio environment are solved.At last, two extended target tracking algorithms based on probability hypothesis density(PHD) were discussed. One of them is based on the random finite set theory, in which the targets and the measurements are modeled by the random finite sets. This approach applies to tracking the non-closed multiple targets in the clutter environment. The other one is a PHD filter based on the random matrix, in which the extent of the target is modeled by a random matrix. This approach applies to track the closed multiple targets in the clutter environment.