Research On Non-line-of-sight Object Reconstruction Based On Single-pixel Imaging

With the development of optical imaging and the continuous increase in application requirements,diversification of imaging methods is evident.As a new target detection and reconstruction technology,non-line-of-sight(NOLS)imaging technology has become one of the most promising and active research directions.Its research results have great application prospects in many fields such as industrial environment hazard monitoring,survivor search in disaster areas,and vehicle positioning and navigation.In NLOS imaging,due to multiple diffuse reflections during the propagation process,the acquired signal amplitude is weak,the spatiotemporal information of the signal is mixed,and the direction information is lost,resulting in failure to image or quality degradation such as reduced brightness,blurred images,and reduced resolution.This thesis uses single-pixel imaging(SPI)technology to perform NLOS imaging.The quality of images and imaging efficiency are improved by the improving image reconstruction algorithms.The research mainly focuses on the following points:1)According to theoretical analysis of NLOS imaging and single-pixel imaging,and the characteristics of NLOS imaging in SPI systems,an active mode/passive mode singlepixel NLOS imaging system is proposed to realize NLOS imaging.Meanwhile,it is verified that single-pixel imaging is feasible in NLOS imaging,and the imaging quality is related to system factors(such as sampling frequency and measurement matrix,etc.)and environmental factors(such as stray light and medium turbidity,etc.).2)A single-pixel NLOS imaging system based on diffraction theory is proposed.Based on the principle of diffraction,the transmission process of light waves is analyzed,and the correlation imaging is combined into the NLOS imaging scene.Random patterns and Hadamard patterns are utilized as illumination patterns.By applying a second-order correlation function,the object is reconstructed quickly,and the influence of diffraction on image quality is reduced by inverse diffraction Different sizes of lenses are used to reconstruct NLOS objects.The analysis and verification in the simulation and experiment demonstrate the practicability of this approach.Besides,compared with the traditional NLOS imaging system,the effectiveness of the system and algorithm for imaging NLOS area targets is verified.3)Aiming at long sampling time in single-pixel NLOS imaging technology,the NLOS object sampling and reconstruction algorithm based on compressed sensing technology is proposed.For binary targets,the Compressed Sensing Orthogonal Matching Pursuit(CSOMP)algorithm is used to obtain high-imaging quality at less sampling time.The experimental results show that a good reconstruction result is obtained with the sampling rate of 29%,and the sampling time and storage space are improved by twice.For the color target,the Total Variation Minimization(TV minimization)algorithm and RGB color space are used to reconstruct the NLOS target.White light is used to modulate the light source,while the receiver utilizes different filters to filter the white light and detects the light information by the sub-channel method.Compared with the traditional correlation imaging method,the proposed system obtains the spatial information and three-channel measurement information of objects at one time with a sampling rate of 29%.The proposed system effectively obtains more information in the same imaging time and reduces the sampling time.4)Aiming at the difficulty in reconstructing objects and poor imaging quality with single-pixel imaging systems in complex environments,a Total Variation Augmented Lagrangian and Alternating Direction Algorithm(TVAL3)is proposed for NLOS transparent object imaging system.The proposed system employs discrete sampling values to achieve spatial information reconstruction of the NLOS transparent object.Experimental and simulation results show that by using the TVAL3 algorithm,clear outlines of transparent objects can be reconstructed with the sampling rate of 19%.In addition,for indirect imaging in the medium,an underwater single-pixel system based on image superresolution is proposed by performing a convolutional neural network.Experiment on the MNIST dataset exhibits the superior performance of the proposed method in underwater imaging compared with other related approaches.The comparison results show that the proposed method can recover good reconstruction results with the sampling rate of 29%.The proposed method shows high robustness against the interference of underwater media,and high-quality images are obtained under low light conditions.Therefore,the proposed method is applicable in imaging objects under complex environments.In summary,this thesis applies single-pixel imaging technology to NLOS imaging,while proposing solutions based on the existing drawbacks of current NOLS object detection and reconstruction.The proposed methods have significant value on application playing an important role in promoting the development of NLOS imaging technology.