Research On Target Recognition Of Ladar Range Images

With the continuous progress of ladar hardware level,the corresponding target recognition technology of ladar range images is becoming a hot research area.Compared with the traditional target recognition technology of intensity images,this technology has many significant advantages.It can be widely used in many fields,such as aircraft navigation,space exploration,sea condition monitoring,precision guided weaponry,remote distance investigation and monitoring.In this thesis,research on target recognition technology of ladar range images is carried out,which can be divided into three parts: the simulation algorithm of ladar range images,the target recognition algorithm of regular ladar range images,and the target recognition algorithm of log-polar ladar range images.The research on target recognition algorithm of regular ladar range images can be divided into two parts: the target recognition algorithm based on the initial slice image and the four improved target recognition algorithms based on the new slice image.In the research on simulation algorithm of ladar range images,a new simulation algorithm of ladar range images is proposed.In this algorithm,the input point clouds come from the Princeton Shape Benchmark,and the position and posture of the input point clouds can be adjusted by using translation vector and rotation matrix,and the noises can be added into range images by using noise models.This system can collect range images with any attitude angle,magnification rate of field of view,object position,image resolution and noise intensity,and not only can it collect range images under the uniform sampling pattern but also collect range images under the log-polar sampling pattern.This research solves a problem that makes very difficult to collect a great number of real ladar range images that meet all experimental requirements.In the research on target recognition algorithm of regular ladar range images,an entire target recognition system of ladar range images is proposed.In this system,a new global feature descriptor called as slice image,a cluster analysis algorithm and a finding the representative slice image algorithm are proposed.It is proven that this feature descriptor meets translation,scaling and approximate rotation invariance.The comparison experiment with the target recognition system based on moment invariances shows that this recognition system meets the properties of high recognition rate,good noise robustness and excellent detail discrimination ability.This research solves a problem on how to represent and classify shape features effectively in the ladar range images.In the research on target recognition algorithm of regular ladar range images,four improved target recognition systems based on the new slice image are proposed.In these four recognition systems,four new global feature descriptors are proposed,and they are new slice image,feature slice image,slice-zernike moment and slice-fourier moment,respectively.The three experiments on three invariances(translation,scaling and rotation),recognition rate and execution time are organized to analyze the performances of these four improved recognition systems.The experiment results show that all the improvements for these four recognition systems reach the desired effect,the three invariances of the feature descriptors are not directly related to the final recognition performance of the recognition systems,and these four improved recognition systems have different performances under different conditions.This research solves a problem on the approximate rotation invariance of slice image and how to improve the recognition performance of entire recognition systems.In the research on target recognition algorithm of log-polar ladar range images,a target recognition system of log-polar ladar range images based on combined moment invariants is proposed.This system needs to transform the log-polar ladar range images into Cartesian ladar range images in advance.An experiment is organized to analyze the change of the three invariances under the log-polar sampling pattern.The other three experiments are also organized to analyze the influence of magnification rate of field of view,object position and noise intensity on the recognition performance of entire recognition systems under the log-polar sampling pattern.The experiment results show that the log-polar sampling pattern brings many new properties to the recognition systems.This research solves a problem on how the log-polar sampling pattern affects the recognition performance of the entire recognition system.