| Non-line-of-sight(NLOS)imaging technology is a technology that breaks through the limit of traditional imaging line of sight and realizes imaging of hidden targets.Through this technology,breakthroughs can be achieved in many scenarios such as medical imaging,autonomous driving,and robotic vision.By measuring the received photons and the corresponding time of flight of the received photon signal after three diffuse reflections from the relay surface--hidden target--relay surface,the 2D or 3D information of the hidden target can be collected to reconstruct the hidden scene.There are many challenges in implementing NLOS imaging.First of all,it consists of the multiple diffuse reflections in the NLOS imaging process and plenty of information is missing.We should assure the accuracy of the received data.The influence of various parameters on the data acquisition and reconstruction results should be accurately analyzed,which has reference for shaping an ideal imaging condition environment.Second,NLOS imaging is a low signal-to-noise ratio problem due to attenuation and environmental noise caused by distance during light transmission,making high-quality reconstruction extremely difficult.How to effectively filter out noise and reduce the influence of inevitable environmental noise on imaging results has practical significance.(1)Based on the problem of complex model of NLOS imaging process,this thesis introduces the three-bounce diffuse reflection attenuation model.A simulation platform for photon counting laser NLOS imaging system is built to realize the full-scene simulation of NLOS imaging.It is shown that the changes of various parameters such as scanning resolution,relay surface roughness coefficient and noise irradiance have a very significant impact on the acquisition of received photons and the reconstruction of hidden targets in this thesis.(2)Based on the problem of low signal-to-noise ratio in the NLOS imaging process,this thesis simulates and models different noise environments by adjusting the noise irradiance value on the simulation platform.The CEEMDAN-ICA method is used for denoising and compared with several other traditional denoising algorithms.Simulation experiments show that the denoising ability of the CEEMDAN-ICA method used in this thesis is better than other algorithms.(3)Based on the problem that mainstream NLOS imaging algorithms reconstruct hidden objects based on the albedo information,which are weak in processing edges and details,an improved light-cone transformation algorithm based on the surface normal vector is used in this thesis.The algorithm combines the surface normal vector with the albedo to achieve the simultaneous reconstruction of the surface normal vector information and albedo information of the hidden target.This algorithm has the same computational complexity as the NLOS imaging algorithm that can only estimate the albedo information of the hidden target,but improves the detail and edge processing capabilities.Compared with the algorithm that can reconstruct the surface contour of the hidden target,the computation time of the algorithm used is greatly reduced. |
