基于深度范围选取的单光子激光雷达高效率图像重建算法

High Photon Efficiency Image Reconstruction Algorithm Based on Depth Range Selection for Single Photon Counting LiDAR

针对在低信噪比、极少量回波光子环境下,三维场景重建能力会受噪声光子及空缺像素影响而下降的问题,提出一种基于深度范围选取的单光子计数激光雷达高光子效率图像重建算法。首先,通过成像场景深度范围选取,去除深度范围外的噪声;然后,采用自适应邻域补充极少回波光子造成的空缺像素以确保场景的连续;最后,使用全变分正则化去除深度范围内残余噪声。通过仿真数据和实验数据验证了所提算法在低信噪比和少光子(小于1)情况下仍能有效重建场景的三维图像。

Photon counting imaging technology is a new type of active imaging technology,which obtains depth information of the target by accumulating histograms of echo photons.It can be combined with Timecorrelated Single-photon Counting(TCSPC)to achieve high temporal resolution.Compared with passive imaging systems,it has stronger robustness and is widely used in fields such as biomedicine,target recognition and remote sensing imaging.But it takes a long time to accumulate thousands of echo photons.In some environments with low Signal-to-background Ratio(SBR)and very few echo photons,such as military reconnaissance and other fields,long-term data collection can not be satisfied,and the ability to reconstruct 3D scenes is affected by noise photons and vacant pixels.This paper proposes a high photon efficiency image reconstruction algorithm based on depth range selection.The algorithm achieves strong resistance to noise and fully improves photon utilization efficiency through two steps:selection of target depth range,adaptive supplementation and TV regularization.Specifically,the selection of the target depth range allows us to gain depth range at the initial stage of the reconstruction process,paving the way for subsequent processing.This process is divided into five steps:merging all data into histogram,peak searching for the histogram,potential signal range determination,signal range review and select signal range.These five processes can ensure that the depth range we obtain is more accurate than setting threshold gating to the histogram.The photon screening process can remove all the noise outside the depth range,thereby reducing the error we introduce when we fill in the vacant pixels.Compared with relying on fixed neighborhood data to supplement vacancies,supplementation using adaptive neighborhood data has a higher photon utilization efficiency and will be more suitable for environments with very few echo photons.Finally,TV regularization is used to smooth the residual noise in the depth range.The simulation and experimental process have verified that even in the case of low SBR and very few echo photons,our algorithm can still effectively reconstruct the 3D image of the scene.We reconstructed the simulation data of different degrees of echo photons when SBR=0.04 and compared it with the high photon efficiency algorithm and the Unmixing method.We also input the data preprocessed by the proposed method into the Unmixing method for processing(Preprocess-unmixing,PP-Unmixing)to verify the contribution of accurately selecting the target depth range.The preprocessing here only includes the selection of target depth range.The results show that our method can distinguish scene edges in any case,and the reconstruction effect and RMSE are better than the other three methods.Our proposed method is also a fast reconstruction method.In addition,a comparison between the PP-Unmixing method and the Unmixing method proves the necessity of accurately selecting the target depth range.In addition to the simulated data,two experimental scenarios further verify the feasibility of the proposed algorithm.In experimental scenario 1,SHIN D′s method can not accurately estimate the depth range,resulting in a large deviation in the reconstructed depth map.And as the number of echo photons decreases,the image becomes increasingly blurred and the scene cannot be resolved,even though it has a fast reconstruction speed.The Unmixing method is better than SHIN D′s method in terms of reconstruction effect,but it still can not completely reconstruct the scene in the case of fewer echo photons,the filtering for noise is not thorough enough,and its running speed is still the slowest.The method proposed in this paper can clearly distinguish the scene in any case.Even in the extreme environment with SPPP=0.47,the time-consuming and RMSE of results are only 0.032 m and 37.2 s.In experimental scenario 2,our method can retain more detailed information than the other two methods,especially in the extreme case when the SPPP=0.7,and the other two methods can hardly detect the house information.Further,in terms of RMSE and time consumption,SHIN D′s method is the fastest,but its RMSE is the largest,and the RMSE of the Unmixing method is comparable to our method,but the reconstruction speed is still the slowest.Therefore,our method has more advantages in comprehensive ability and is more suitable for an environment with few echo photons and low SBR.In summary,our method has a significant reconstruction effect on both simulation data and experimental data,which proves that this method is more suitable for the situation of extremely low SBR and a very small number of echo photons.In terms of computing speed,it is also a fast reconstruction method.In addition,the proposed method has better applicability to the situation where there are multiple depth targets in the scene,and further research and verification will be carried out in the future.