Study Of Typical Three-dimensional Ground Target Recognition Based On Laser Radar

Imaging laser radar (LIDAR) is a kind of high precision imaging sensor with high range resolution, high angular resolution and high measurement accuracy. The study of 3-D automatic target recognition technology based on laser radar has important significance on military and theoretical research. In this paper, 3-D target recognition based on spin-image is studied. In the research, the background is imaging laser radar, the experimental data is from laser radar simulation, the feature of target is spin-images, and the matching algorithm is similarity measure. The main contents of this study are as follows:(1) In the 3-D target data acquisition and processing, firstly, we specifically analyze the computing process of intersection coordinates between the laser beam and the scene for ground data and the loaded 3-D model; secondly, we discuss the expression of the LIDAR data and choose irregular triangular mesh which is fit for spin-images algorithms to describe 3-D model; at last, point cloud target extraction method and 3-D surface reconstruction method is proposed.(2) In spin-images based recognition algorithm, the article describes spin-images generation process and proposes four identification criterias for scene target and model target. The four criterias are combined, constituting a combined recognition criteria. The combined recognition criteria can be adaptively adjusted for different imaging conditions, to achieve balance between performance and efficiency. The main idea of the target recognition algorithm is: after the 3-D imaging LIDAR point cloud are transformed into 2-D space of spin image, then the problem of 3-D target recognition is converted to matching problem between 2-D model spin-images and scene spin-images.(3) In target motion mode, the three-dimensional imaging model and the type of geometric distrotion are analyzed. The relationship between geometric distrotion and target motion parameters and the effects of geometric distrotion on feature extraction is derived.(4) In three-dimensional target recognition experiment, we first set up the database of typical ground targets; and then we study the influence of target spatial resolution, LIDAR ranging noise and target occlusion on the target recognition. The experimental results demonstrate that the proposed recognition method is feasible. The study can provide reference for parameter design of the 3-D target recognition system.