Fast Surface Detection Using Single-Photon Detection Events

This paper presents a fast object detection algorithm for 3D single-photon Lidar data. Lidar imaging acquires time-of-flight (ToFs) events in different spatial locations to build a 3D image of the observed objects. However, high ambient light or obscurants, might affect the reconstruction quality of the 3D scene. This paper proposes a solution by first detecting the pixels containing photons reflected from a object/surface, allowing a higher level processing of the data while only accounting for informative pixels. In contrast to histogram based approaches, the proposed algorithm operates on the detected photon events allowing a reduction in memory requirements and computational times. A Bayesian approach is considered leading to analytical estimates that can be computed efficiently. Results on simulated and real data highlight the benefit of the proposed approach when compared to a state-of-the-art algorithm based on histogram of counts.