Semi-automatic Integration of Panoramic Hyperspectral Imagery with Photorealistic Lidar Models

This paper presents a method for increasing the automation of the registration of panoramic hyperspectral images with co-registered conventional digital imagery and a point cloud acquired with a terrestrial laser scanner (lidar), for geological purposes. The SIFT (scale invariant feature transform) interest operator is used to find homologous points between the two imagery types which, because they are recorded at different spectral ranges (visible and short wave infrared) using different geometric projections, differ significantly in appearance. After reducing false matches using RANSAC (random sample consensus), a geometric model for panoramic cameras is applied to retrieve the orientation parameters of the hyperspectral scenes. Once registered, hyperspectral classifications can be combined with the lidar geometry in photorealistic models, allowing material information to be linked to object geometry. Improved automation of the data registration reduces processing time and makes the domain more accessible for non-specialists.