Fast Classification and Depth Estimation for Multispectral Single-Photon LiDAR Data

Multispectral 3D LiDAR imaging plays an important role in the remote sensing community as it can provide rich spectral and depth information from targets. This paper proposes a fast pixel-wise classification algorithm for multispectral single-photon LiDAR imaging. The algorithm allows the detection of histograms containing surfaces with specific spectral signatures (i.e., specific materials) and discarding those histograms without reflective surfaces. The proposed Bayesian model is carefully built to allow the marginalization of latent variables leading to a tractable formulation and fast estimation of the parameters of interest, together with their uncertainties. Results on simulated and real single-photon data illustrates the robustness and good performance of this approach.