| Active imaging lidar systems use pulsed lasers and detectors to recover useful scene information such as the three-dimensional structure,reflectivity or fluorescence of an object.A large number of photons are typically required to form the final image to satisfy 100 to 1000 photons per pixel to generate a very fine histogram,and then the histogram is analyzed to extract the depth and reflectivity information of the scene.Due to the Poisson noise inherent in the photon counting process of a single photon detector,the histogram estimation is greatly affected by the noise level at this time,so it is necessary to find a new imaging method to achieve accurate three-dimensional reconstruction.In the low light-level scenarios,the photon time of flight data collected by the detector also contains a lot of noise from background and detector,so how to separate the signal from noise,and extract useful information from it becomes the key to three-dimensional reconstruction of lidar.Firstly,since the signal detections tend to cluster together more readily in the time domain,and the noise detections are distributed randomly,a time correlation is established in combination with the flying photon.The windowing approach in our proposed algorithm is used to distinguish between the signal and noise in the time domain.The first experimental results show that under the different signal-to-background ratio(SBR)conditions,this method has a good performance in 3D reconstruction for the scene.The root mean square error(RMSE)is reduced by 62% compared to the traditional maximum likelihood(ML)estimation method.Secondly,since signal photons may not detected in some pixels,such time domain methods cannot be used to extract useful information.Therefore,considering the spatial correlation of the scene,another improved algorithm is proposed to realize the three-dimensional reconstruction of the point of interest by combining the signal photon estimation values from fixed-distance similar pixels.The second experimental results show that under the condition of SBR=10,the accuracy of this algorithm is about 2.2 times higher than the ML algorithm,and under the condition of SBR=1,the accuracy of this algorithm is 9.6 times higher than the ML algorithm at most.The algorithm is not only demonstrated the excellent performance,but also can be applied to a variety of complex scenes,making three-dimensional reconstruction of multiple depth targets accurately. |
