Research On None-line-of-sight 3D Imaging Based On Optimization Method

Non-line-of-sight three-dimensional imaging is a technology that uses multiple scattered light to image non-line-of-sight targets.By using an ultra-short pulse laser and a single photon counter with ultra-high time resolution to obtain the time-of-flight information of multiple scattered light in a non-line-of-sight scene,three-dimensional image reconstruction of the non-line-of-sight target can be achieved.As a rapidly developing imaging technology in recent years,it has broad application prospects in many fields such as medical imaging,machine vision,anti-terrorism investigation,and unmanned driving.Traditional non-line-of-sight imaging method: The reverse ellipsoid algorithm,by calculating the time of flight of the three scattered photons at different imaging points in space,and then draws ellipsoids in space with the illumination point and imaging point as the focal point.The ellipsoids drawn at different imaging points are superimposed to finally obtain the photon number distribution in space,thereby obtaining the target image.The result of reverse ellipsoid algorithm always has image artifacts,and the image quality is low.This dissertation proposes that by establishing a light field transmission model in a non-line-of-sight scene,the image restoration process of non-line-of-sight imaging can be converted into solving a constrained objective function.Through establishingt the model,the light field transmission process is quantified to generate the light field transmission matrix,and the measurement result vector is generated based on the TCSPC data.With the help of the optimization restoration algorithm,we can directly solve the objective function to complete the reconstruction of the target's high-quality image.This dissertation introduces the principle of the reverse ellipsoid algorithm,establishes non-line-of-sight light field transmission model,introduces and uses three optimization restoration algorithms,including the fast gradient descent method using sparse constraints,the Split-Bregman method using tatal variation constraints,and the popular algorithm alternating direction multiplier method using sparse constraints.First,the simulation research was carried out through the knowledge of ray tracing and geometric optics.The image restoration results under different sig nal broadening conditions were compared.The simulation studies of no-noise and noisy conditions were carried out,and the comparison of different algorithms with different iterations was initially compared.The results show that the gradient descent method is the fastest,20 iterations take about 10 s,the alternating direction multiplier method is the slowest,20 iterations take about 600 s,the Split Bregman method is centered,and 20 iterations take time About 30 s,the image restoration result is the best and the target boundary is the clearest.Using a picosecond pulse laser and a single photon counting array,a non-line-of-sight imaging experimental system was built,experimental research was conducted,non-line-of-sight imaging experimental data was obtained,The non-line-of-sight experimental image restoration based on reverse ellipsoid algorithm and optimization method was carried out.The results show that under the same spatial resolution,the image restoration result of the optimization method is better,and the resolution of the restored image is about 10 cm.Finally,the three algorithms used in this dissertation are studied and explored.Using the control variable method,the influence of each algorithm's parameters on the imaging results is analyzed,and the parameter setting guidance is provided for obtaining higher-quality restored images.