Robust depth imaging in adverse scenarios using single-photon Lidar and beta-divergences

This paper addresses the problem of robust estimation of range profiles from single-photon Lidar waveforms associated with single surfaces using a simple model. In contrast to existing methods explicitly modeling nuisance photon detection events, the observation model considered neglects such events and the depth parameters are instead estimated using a cost function which is robust to model mismatch. More precisely, the family of $\beta -$divergences is considered instead of the classical likelihood function. This reformulation allows the weights of the observations to be balanced depending on the amount of robustness required. The performance of our approach is assessed through a series of experiments using synthetic data under different observation scenarios. The obtained results demonstrate a significant improvement of the robustness of the estimation compared to state-of-the-art pixelwise methods, for different background illumination and imaging scenarios.