Converting non-confocal measurements into semi-confocal ones with timing-accuracy improving for non-line-of-sight imaging

Non-line-of-sight (NLOS) imaging enables an imaging system to look around corners, which has drawn much attention in recent years. With the confocal setup employing the raster scanning of overlapped illumination and detection points on the relay wall, NLOS imaging has well-established image reconstruction algorithms. In order to elevate the detection efficiency, the single-photon-avalanche-diode (SPAD) array is actively developed and has become a well acknowledged way to improve the practicability of NLOS imaging. However, applying SPAD array naturally leads to the non-confocal configuration (illumination and detection points not overlapped) and lacks efficient image retrieving algorithms, which remains a serious problem in NLOS imaging. In this work, non-confocal measurements obtained by point illumination and SPAD-array detection are successfully transformed to semi-confocal measurements using the proposed approach based on sectionalized ellipsoid interpolation (SEI), in which the NLOS object is represented as multiple ellipsoid sections corresponding to the non-confocal histogram. By investigating the fact that the ellipsoid section can be refinedly sliced and each slice directly maps to each time-bin in the new confocal histogram, the timing resolution for the emulated confocal measurements can be improved even beyond that for the detector itself. Simulations and experiments have been conducted and testified the advantage of the proposed SEI-based approach over the existing filtered backprojection (FBP) and normal-moveout-correction-based (NMO-based) approaches. Bar-shaped objects (used for resolution estimation) with the interval of 65.6 mm in simulation and 100 mm in experiments, which are not resolved by FBP and NMO-based approaches, can be distinguished by the SEI-based approach with the image contrast of 0.094 and 0.228, respectively. For a large E-shape object with the horizontal size of 1.18 m in simulation, the reconstruction nonuniformity is reduced to 0.169 by the SEI-based approach with respect to 0.271 with the NMO-based approach. Confocal measurements emulation using SEI successfully connects the challenging non-confocal NLOS problem to the well-established confocal NLOS reconstruction algorithms with high performances, which is especially meaningful to NLOS imaging using SPAD array with bad timing resolution and contributes to the practical applications of NLOS imaging.