| When observing,both human eyes and conventional imaging equipment can be obstructed by some shelters or obstacles.The viewing area is limited and there is a certain visual blind area.In scenes such as rescue,counter-terrorism,and investigation,the observation of non-line-of-sight(NLOS)space is crucial,so NLOS imaging technology has received more and more attention.With the development of highresolution detection technology and the relative algorithm research,NLOS imaging using laser becomes possible,and more and more scholars have conducted related research.Based on the in-depth understanding of the research status at home and abroad,this thesis focuses on the energy transfer model in the NLOS imaging process,selects and models devices for the experimental scene,and simulates the laser NLOS imaging process.Based on the simulation data,an inversion and reconstruction was performed and the influence of various factors on the imaging was analyzed:1)Introduce relevant theories of laser NLOS imaging simulation,including energy transmission model and ray tracing during the imaging process,the detection,sampling and quantification method of the detector,and inversion algorithm and surface reconstruction theory after data acquisition.The resolution in NLOS imaging was analyzed in detail,and the advantages of resolution imaging was pointed out.2)Set up NLOS scenes with wall obstructions,load pictures to certain position as 2D target and read 3D models as 3D target,use the improved point cloud tracking method to track the targets’ light propagation process.Set models for key devices,simulate the data acquisition process,and calculate echo distribution of the scene.The simulation data was collected for later inversion and reconstruction and the idea of simulation can play a certain role in actual system setup.3)The classical ellipsoid inversion algorithm is used to invert the target space based on the simulated data.Through the grid division,data matching,grid probability assignment,and screening and contour extraction,the target area is obtained,and then the target imaging map is drawn.Imaging results of 512 ×512 2D targets show that with proper parameters,the ellipsoid inversion method can be used to observe 2D targets with few details,for example imaging result of the picture XIOPM reaches a structure similarity(SSIM)of more than 0.9 with the original target.But this method is more demanding on the scene and can image 2D objects under certain conditions,but the imaging results of 3D objects are not recognizable.4)Light cone transformation(LCT)inversion is realized.LCT starts from the imaging process.By the method of Wiener filter,the reflectivity of the NLOS space is approximated by iterative inversion.The time resolution,space resolution and computation are related.LCT inversion has a strong tolerance for targets,can obtain quite good imaging result of both 2D and 3D targets,for example imaging result of the 256×256 XIOPM reaches an SSIM of more than 0.5 with the original target and the projection of the imaging of 3D bunny has an SSIM of 0.54533 with that of the original target.5)Three kinds of surface reconstruction algorithms,Poisson,greedy triangular projection and powercrust,were studied.The same point cloud was used for comparison experiments.Reconstruction of a series of bunny using different methods was achieved,and the reconstruction results of several methods were compared.Among them Poisson reconstruction is suitable for closed-surface reconstruction with determined normal vector,greedy triangular projection reconstruction method is suitable for relatively uniform point clouds,and powercrust is most suitable for the reconstruction of scattered point clouds.Use powercrust to reconstruct the LCT inversion result of bunny,a complete surface was obtained and the visual effect was optimized.The research results show that it is feasible to use time-resolved methods to observe NLOS objects.In certain scenes,2D targets’ images can be obtained by ellipsoid inversion,but this method cannot be used to observe 3D objects.LCT inversion can achieve better results than ellipsoid inversion and get an imaging result close to the original target,but the system setup is more difficult.The results of imaging can be affected by many factors and this imaging method has high requirements on the system,scene and target.There is still a long distance away from actual application. |
