An image feature-based approach to improving SPAD flash LiDAR imaging through fog

Fog is a difficult medium to image through using Single-Photon Avalanche Diode (SPAD) based Light Detection and Ranging (LiDAR) systems because of its light scattering properties. Scattering significantly decreases the signal-to-noise ratio of photon returns, making it difficult to reconstruct meaningful images for target detection. In this paper, an image feature-based approach for reconstructing SPAD LiDAR images of a single target is proposed. Geometric characteristics of the target are used in the algorithm to differentiate between target and background photon returns. Combinations of different features such as Fourier shape descriptors and apparent target size are used to improve performance. To validate the algorithm, a 32x32 silicon SPAD array Flash LiDAR system operating at 532nm is used for collecting images through fog. Simple geometric shapes are placed indoors in a dark tunnel 44.6m from the sensor with fog decreasing the visibility in steps down to 12m. The proof-of-concept algorithm achieves good localization performance at a fog level of 1.4 attenuation lengths.