An Empirical Approach on Shadow Reduction of UAV Imagery in Forests

This work is aimed at the recovery of shadowed pixels in optical imagery acquired from Unmanned Aerial Vehicles (UAV) for environmental purposes. When acquiring imagery with UAV, scene dependent elements as trees, man-made constructions or the terrain morphology can block the sun beam of light (direct irradiance) and to project shadows, causing the surface to only receive diffuse light (diffuse irradiance). We present a method to reduce the effects of the shadows and to recover information based on in-situ spectral reflectance measurements. The shadow reduction is done using a pair of spectrally characterized target sets, one set placed under direct irradiance (non-shadowed targets) and, the second set, with the same spectral characteristics under diffuse irradiance (shadowed targets). Assuming that the target reflectance is the same in both sets located in the shadows and in direct illumination, we model the image values in shadowed targets using the image values of non-shadowed targets and the model is applied to all shadowed areas. The method can be very useful in forested areas sensed by drones (e.g., in detailed new forests identification, in protected areas monitoring, etc).