Research On Large View SAR Image Registration And Low-speed Group Targets Tracking Under Moving Platform

With the increase of resolution and detection performance, wide-area surveillance and tracking by modern radar is available. SAR image registration is a basis procedure in wide-area surveillance. Severe geometry distortion and diverse resolution caused by large variation in view angel as well as lots of speckles result in the poor performance of registration between large view SAR images. Group targets tracking plays a vitally important role in informationalized defense. High-resolution radar has the capacity of detecting measurements produced by group targets in low speed, but it's positioning error and systematic error lead to disordered distributions that makes the tracking precision of each target reduced. This thesis mainly deals with the algorithm for SAR image registration and tracking of group targets moving in low-speed. The main contents include:1. Classical algorithms for optical image including knowledge of the gray level based and characteristics based which is the basic of SAR image registration are studied, while SIFT algorithm is mainly discussed.2. Point target tracking framework which contains initial algorithms, data association algorithms and filtering algorithms is introduced for the sake of following research that focus on tracking of group targets which are considered as a multi-point-targets group.3. A large-view SAR image registration algorithm based on feature enhancement is presented. This algorithm firstly extracts closed-boundary regions which are described by affine moment invariants as the features of image; secondly enhances the features using motion and imaging parameters, which called preprocessing; lastly obtains the transformation matrix between converted image and reference image by solving the parameters between converted image and preprocessed image. Match precision is increased by feature enhancement meanwhile time-consuming is reduced by feature extraction and description. Performance of presented algorithm is confirmed by contrasting with SIFT.4. A method based on the shape description of group targets is presented. This approach firstly gets the partition of multi-targets; secondly constructs a descriptor using the distance and angel between each target and group center; thirdly does data association of each track in the group, which is controlled by the group shape; lastly updates the shape using filtered state of each track. Processing efficiency is increased by measurement distribution of group members based on shape description which avoids repeated association and intersected association. High performance of presented method is confirmed by contrasting with classical multi targets tracking algorithm.