Research Of Non-Line-of-Sight Imaging Based On Time-of-Flight

Telescopic and microscopic techniques expand human visual ability,but the traditional imaging technology is still unable to break through the limitations of obstacles.The proposal of non-line-of-sight imaging makes it possible to observe the scene behind obstacles.Non-line-of-sight imaging uses the diffuse reflection signals returned from relay surfaces such as walls and curtains in the field of view to reconstruct the image.Based on the theory of time-of-flight detection and counting,we studies the image quality and reconstruction algorithm of non-line-of-sight imaging from simulation and experiment.The direct intersection of ellipsoid and voxel is combined with Gram matrix optimization to improve the reconstruction speed while maintaining a certain reconstruction quality.The diffuse reflection light transmission model of non-line-of-sight imaging is derived,the time-of-flight imaging technology and time-dependent single-photon counting technology are studied,several ultra-high time-resolution detectors are introduced,and the non-line-of-sight imaging hardware system is designed.A non-line-of-sight imaging system composed of ultrafast pulse laser,scanning galvanometer,ultra-high time-resolution detector and time-dependent single-photon counting module is designed,and a superposition reconstruction algorithm is developed to distinguish the intersection of ellipsoid model and spatial voxels.The reconstruction speed is improved.Under the qualitative condition,the influence of sampling parameters on the reconstruction quality is explored.The study shows that when other conditions are constant,the reconstructed image will be distorted with the offset of the sampling point,and the reconstruction quality can be improved with the increase of sampling density and range.However,after reaching a certain extent,the reconstruction accuracy will no longer change significantly,and there is a quantitative relationship between longitudinal depth and sampling range.The study provides a reasonable sampling point setting scheme for the experiment.A non-line-of-sight imaging experimental system is built,and the image of the plane object is reconstructed by using silicon photomultiplier tube and single-photon avalanche photodiode respectively.according to the analysis of the reconstruction results: the silicon photoelectric multiplier tube needs to set an appropriate voltage threshold to select the single-photon signal of each group of sampling points,which affects the time-of-flight detection accuracy,resulting in poor reconstruction quality.The reconstruction algorithm is optimized according to the Gauss Laplace filter and the least square problem of convolution approximation of Gram matrix,respectively,and the three scenes of plane object,object with certain depth and double target object are reconstructed.the results show that after the optimization of the algorithm,the geometric shape of the reconstructed image is more significant,and the intensity of the depth object reconstruction image decreases with the increase of depth.It is found that the lateral resolution accuracy of the non-line-of-sight imaging system is higher than 3cm,and the reconstruction effect of the algorithm based on Gram matrix optimization is better than that of Gauss Laplacian filter.