Research On Passive Non-line-of-sight Image Restoration Method

In recent years,with the development of visual imaging equipment and image restoration algorithm,non-line-of-sight imaging has become a new research field.At present,more and more researchers at home and abroad have invested in the field of non-line-of-sight imaging.This technology has a broad application prospect in the fields of unmanned driving,anti-terrorism operations,disaster relief,service robots and so on.However,in the current research,there is a lack of means to improve the imaging quality in non-line-of-sight imaging methods.In order to solve the above problems,this paper focuses on the passive non-line-of-sight image restoration method.The main research contents are as follows:(1)For the problem of image degradation in the process of non-line-of-sight optical transmission,the passive non-horizon visible image restoration algorithm is adopted.The advantages of passive non-line-of-sight imaging compared with active non-horizon imaging are explored,and the experimental environment of passive nonline-of-sight imaging containing occlusion is introduced.According to the experimental environment conditions,combined with the image restoration theory and the principle of non-line-of-sight optical transmission,the discrete non-line-of-sight image degradation function is constructed.Finally,a passive non-line-of-sight visible light imaging scene is built to carry out the basic experiment of image restoration,and the corresponding results are obtained.(2)Aiming at the problems of high-frequency noise and low contrast in the passive non-line-of-sight visible image restoration results,a non-line-of-sight visible image restoration algorithm based on the bidirectional reflectance distribution function model is proposed.The mirror reflection part that is not considered by the original degradation function is modeled by the bidirectional reflection distribution model,and the corresponding non-line-of-sight visible light image restoration method is designed,and the corresponding experiments are carried out.The experimental results show that the restoration algorithm based on bidirectional reflection distribution model can obtain better imaging effect than the original algorithm,and eliminate some high frequency noise blocks,and improve the relevant image restoration index.In addition,sensitivity analysis is carried out for the relevant parameters of different mirror reflection proportions and different bidirectional reflection distribution models to obtain the influence law of non-line-of-sight image restoration parameters.(3)Aiming at the problem that the temperature difference between target and environment is not obvious in the current non-field infrared image restoration method,a Uformer non-line-of-sight infrared image restoration algorithm based on MSSSIM+L1 loss is proposed.The experimental scene of non-line-of-sight infrared image restoration was set up,and a data set containing 15 kinds of numbers and letters of1200 infrared images was produced for corresponding experiments.The experimental results show that Uformer can intuitively achieve better non-line-of-sight infrared image restoration results than U-net,and its image restoration parameters are better than U-net.Through ablation experiments,the influence of core components in each part of Uformer on the computational power and image restoration quality of the model was analyzed.In addition,the MS-SSIM+L1 combined loss function is proposed to replace the original loss function.The experimental results show that the proposed method can reduce the training loss,further improve the intuitive effect of non-lineof-sight infrared image restoration,and significantly improve the parameters of image restoration.