Research On Key Technologies Of Laser 3D Imaging

With the development of laser imaging technology,3D laser imaging radar,owing to its own unique advantages,is playing an increasingly important role in modern military field.Based on the characteristics of imaging data acquired by self-developed 3D imaging laser radar,this thesis focuses on the core issue of 3D visualization and recognition of objects,and researches on related work including modeling and simulation of 3D laser imaging,surface smoothing,surface reconstruction,3D feature extraction and target recognition.The specific research contents and innovations are as follows:(1)Research on simulation technology of 3D laser imagingAfter study on the theory model of 3D laser imaging,we research on simulation of 3D laser imaging,and develop two schemes to simulate the physical interaction logic between laser beam and target surface.With simulation schemes developed by us,ideal 3D imaging result of targets can be simulated,which will help to validate and discuss on the design of 3D imaging system in the stage of system design,and facilitate research in advance on data processing algorithm of 3D laser imaging radar,therefore help to reduce the error cost and shorten the development cycle of the whole system.The geometric simulation method creatively introduces raytracing into the laser 3D imaging simulation,and performs a local search,so it can efficiently simulate the 3D imaging process of the laser 3D imaging radar to any target.Based on the proposed simulation schemes,we simulate the ideal imaging process of 3D laser imaging radar developed by NUDT to typical aerial target in different spatial positions with different state of motion.And based on those simulation results,we carry out researches in following algorithms.(2)Research on surface smoothing algorithmSurface smoothing is an important preprocessing in the reconstruction of target's surface,which can reduce the influence of noise on the measured data and make reconstructed surface smooth.The most popular algorithm of surface smoothing is upon the idea of bilateral filtering,its core idea is adjusting vertex position along the direction of normal vector to achieve surface smoothing,and that the size and direction of position adjustment depend on the triangular patches in the neighborhood of the vertex to be adjusted.However,when the sampling of the laser is sparse and the calculation error of normal vector of vertex is relatively large,such surface smoothing algorithm cannot work well any longer.We make full use of the spatial topological information contained in the range image of laser 3D imaging system,and give full consideration to the limitations of 3D bilateral filtering algorithm when applied to sparse point cloud,then carry out the two dimensional processing with a combination of image median filtering and bilateral filtering,so the isolated noise points and the small amplitude noise in the image are filtered out with maintaining surface's shape to the maximum extent.Experimental results show that the proposed method can effectively make surface smooth,and its effect is better than traditional algorithms',furthermore it is 8 times faster than traditional algorithms,so it can meet the real-time requirements more.The proposed surface smoothing algorithm,whose physical meaning is adjusting vertex's position along the direction of emitting laser to achieve surface smooth,not only makes full use of the advantages of bilateral filtering in surface smoothing,but also avoids the calculation of vertex's normal vector,therefore smoothing effect and speed can been greatly improved.(3)Research on surface reconstruction algorithmThe laser 3D imaging radar obtains the sampling point cloud on the target surface.Point cloud itself does not contain any spatial topological information,so it cannot well represent target's geometric shape,and cannot well provide the real stereo perception either.Therefore achieving surface reconstruction based on point cloud,is the core content of 3D visualization in laser 3D imaging radar.Currently,piecewise reconstruction is the mainstream method in surface reconstruction,it generating triangular patches through space topology on discrete point cloud,and then piecewise constructing mesh model with boundary conditions of vertices' coordinates and normal vectors,finally achieving surface reconstruction.But without surface segmentation and merging,the traditional piecewise reconstruction cannot be applied to reconstruct objects with sharp edges.Given that,we propose a method for mesh piecewise reconstruction based on feature detection,which can directly reconstruct targets with sharp edges,and can be applied to reconstruct targets in laser scanning system.Firstly,the method builds the target's triangular grid model;secondly it performs an operation of feature detection on the grid model,and then carries out Kmeans clustering on the triangle patches in the neighborhood of key point,estimates key point's normal verctor based on the weighted average of normal vectors of neighborhood triangle patches of each clustering,obtains the normal vector set of each grid vertex;finally,with the information of vertices' positions and normal vector sets,it implements a quadratic interpolation on each triangle patch,so the 3D reconstruction of the target is achieved.The proposed method,avoiding the complicate segmentation and merging,is quite simple.The experimental results show that it can make C0 continuity attained,so it meets the need of visualization.Moreover,the reconstruction speed can be greatly improved compared with the traditional algorithm because of the avoidance of the surface segmentation preprocessing and the operation of the surface splicing.(4)Research on feature extraction and recognition of 3D targetsBased on 3D imaging data,to distinguish fighter planes from aircrafts,is an important step to accurately identify suspected targets in the air.According to the characteristics of the imaging data of air targets acquired by self-developed 3D imaging laser radar,seeking for 3D feature description method and recognition algorithm for a typical aircraft target,is an urgent issue in the field of military defense.Base on the theory of covariance matrix and PCA signal processing,we propose an efficient estimation method of 3D OBB of point cloud,and extract geometry information from the 3D OBB as target's feature descriptor to help to distinguish fighter planes from aircrafts accurately.Experimental results show that the proposed algorithm based on 3D OBB and linear SVM classifier can achieve robust classification of fighter and aircraft,with 10% higher classification rate than that based on Hu moments of range image and 15 times faster than that based on Zernike moments of range image.As a fast and effective algorithm for aircraft recognition,the proposed algorithm can be applied in laser imaging radar especially developed independently at home to improve combat capacity.