Efficient joint surface detection and depth estimation of single-photon Lidar data using assumed density filtering

This paper addresses the problem of efficient single-photon Lidar (SPL) data processing for fast 3D scene reconstruction. Traditional methods for 3D ranging from Lidar data construct a histogram of the time of arrival (ToA) values of photon detection events to obtain final depth estimates for a desired target. However processing large histogram data volumes over long temporal sequences results in undesirable costs in memory requirement and computational time. By adopting a Bayesian formalism, we combine the online estimation strategy of Assumed Density Filtering (ADF) with joint surface detection and depth estimation methods to eventually process SPL data on-chip without the need for histogram data construction. We also illustrate how the data processing efficiency can be increased by reducing the set of unknown discrete variables based on posterior distribution estimates after each detection event, reducing computational cost for future detection events. The benefits of the proposed methods are illustrated using synthetic and real SPL data for targets at up to 3 km