Research On Super-resolution 3D Reconstruction Algorithm Of Laser Composite Imaging With Different Resolution

Gm-APD lidar has become a hot research field today due to its high detection sensitivity,high range resolution,and easy integration.However,its spatial resolution is usually low,and the ICCD lidar intensity image has high spatial resolution.In order to improve the spatial resolution of Gm-APD range image,a dual wavelength composite imaging system of Gm-APD lidar and ICCD lidar is built to collect low resolution range image and high resolution intensity image,and a region similarity guided super-resolution three-dimensional reconstruction algorithm is studied.The algorithm uses the advantage of high spatial resolution of ICCD intensity image to guide the low spatial resolution Gm-APD range image to generate high spatial resolution Gm-APD range image.Aiming at the two shortcomings of the algorithm,which are uncontrollable standard deviation of area similarity guide term and fuzzy edge of reconstructed image,this paper proposes an improved image guided super-resolution 3D reconstruction algorithm.First,the research status of image super-resolution 3D reconstruction algorithm is investigated.Through analysis of the research status at home and abroad,the image guidance algorithm guided by regional similarity is taken as the main research direction of this article.Therefore,this paper studies the principle of image guidance algorithm guided by regional similarity,obtains the analytical solution of the optimization objective equation of the algorithm through detailed formula derivation,and studies the evaluation indexes of non reference image quality and reference image quality to evaluate the quality of super-resolution reconstructed image.Further use bicubic interpolation,standard image guidance algorithm,guidance filtering,TGV and regional similarity guidance algorithm to perform super-resolution reconstruction processing on simulation data and real radar data,and use subjective visual effects and image quality evaluation index to evaluate the performance of each algorithm.The advantages and two disadvantages of the regional similarity guidance algorithm are analyzed.The advantage lies in the research on the results of simulation data.It is found that the reconstruction image of the region similarity guidance algorithm is the clearest,the edge is the sharpest,and the phenomenon of texture replication is less.It obtains the best subjective visual effect and the best index performance.The disadvantage lies in the uncontrollable standard deviation of the region similarity guidance item,which leads to the instability of reconstruction quality,and the edge of the reconstructed image is fuzzy.Secondly,in order to solve the problem of edge blur of reconstructed image,this paper proposes four algorithm improvements: adaptive standard deviation of Gaussian kernel function,norm model construction based on local content perception,high resolution distance guided image based on regional interpolation and edge penalty guided item of super-pixel segmentation;To solve the problem of uncontrollable standard deviation of region similarity guide,this paper proposes a region similarity guide of controllable Gaussian kernel function.In this paper,the iterative weighted least square method is used to solve the optimal solution of the improved algorithm,and the iterative analytical solution of the optimal equation is obtained by detailed formula derivation.Further,the simulation data is used to verify the performance of each improved algorithm by subjective visual effect and objective evaluation index of reference image quality,and the simulation data is super-resolution reconstructed by bicubic interpolation,guidance filtering,TGV,standard image guidance algorithm,region similarity guidance algorithm and all improved image guidance algorithms.The results show that the improved algorithm has higher resolution,sharper edge,less texture duplication and the best performance.Compared with other algorithms in this paper,RMSE is increased by up to 72%,and SSIM is increased by up to 7%.Finally,a dual wavelength lidar composite imaging system is built,and low resolution Gm-APD range image and high resolution ICCD intensity image are obtained by preprocessing the image collected by the imaging system.Further,the radar image is used to verify the performance of each improved algorithm by subjective visual effect and objective evaluation index of unreferenced image quality,and the radar image is super-resolution reconstructed by bicubic interpolation,guidance filtering,TGV,standard image guidance algorithm,region similarity guidance algorithm and all improved image guidance algorithms.The results show that the edge of the reconstruction image of the improved algorithm is sharper and the index performance is better than other algorithms.Compared with other algorithms,the maximum Brenner gradient is increased by 151%,the maximum energy gradient is increased by 357%,the maximum Laplacian gradient is increased by 31%,and the improvement range is obvious.