Registration Technology For LADAR Range Image

Since the environment of battlefield is becoming more and more complex, therequirements for precision guided weapons are becoming even challenging. Laser Radar(LADAR) is active sensor which can directly acquire the three dimensional (3D)information of a target. It is expected to be an up-and coming technique for detectingand recognizing ground targets in the presence of complex background. One importantresearch direction for precision guidance techniques is the laser imaging precisionguidance. LADAR can produce both intensity and range images, it provides moreinformation (especially3D information) for target recognition. Therefore, LADAR issuperior to infrared imaging sensors and Synthetic Aperture Radar for target detection.However, the data of a3D scene (target) acquired with a LADAR is not exactly thesame as the real scene. In order to get the real characteristics of a scene (target) andprovide more appropriate data for feature extraction and target recognition, it isimportant to perform preprocessing on the raw data. This paper works on LADARrange image registration, the main contents are listed as follows:The first chapter works on feature matching based point-cloud registration method.It firstly introduces the popularly used registration algorithm–Iterative Closest Point(ICP) algorithm in details. It analyzes the principle and each step of the ICP algorithmin details. Coarse registration is an effective solution to solve the problem faced by theICP algorithm. Feature matching based method is a frequently used coarse registrationmethod. Thus chapter introduces the principle of the proposed method, and theproblems of feature point extraction and feature point matching. The effectiveness ofthis method is finally demonstrated on a synthetic LADAR dataset.The second chapter works on least consistent surrounding rectangle (LCSR) basespoint-cloud registration. Considering the particular characteristics of ground targetpoint-clouds, a novel method (i.e., LCSR method) is proposed. It adopts LCSR todescribe the contour information of the target point-cloud on each coordinate plane. Itestimates the independent target-centered coordinate system by finding the minimumsum of the areas of bounding rectangles on all the coordinate planes. It then performscoarse registration by aligning the independent coordinate systems of all range imagesto the world coordinate system. It finally uses the ICP algorithm to perform fineregistration. The proposed method is tested on a synthetic dataset, with comparison tothe feature matching based point-cloud registration method.The fourth chapter works on iterative least spatial distribution entropy (ILSDE)based point-cloud registration method. It studies the relationship between the spatialdistribution and the relative position of a point-cloud, and proposes a spatial distributionentropy based on the concept of “entropy” used in the information theory. It then analyzes the relationship between the spatial distribution entropy and the spatialdistribution of a point-cloud. Based on these analyses, it proposes an ILSDE basedpoint-cloud registration method. It finally demonstrated the proposed method on asynthetic dataset with rigorous comparison to the feature matching based registrationmethod.The last chapter is a conclusion of this thesis, which sums up the main points of thethesis and presents some advises on advanced research.