Toyon has developed software that performs automated 3D model reconstruction using multiple overhead images, e.g., from satellites or aircraft. The software implements a novel Bayesian Multi-View Stereo (BMVS) algorithm developed by Toyon researchers. The algorithm performs near-optimal processing of the available data to reconstruct a fused 3D model with minimum final uncertainty. The algorithm effectively addresses challenges due to low-texture scenes, coarse measurement resolutions, and/or low signal-to-noise ratios that may occur in remote sensing applications. Furthermore, the algorithm accounts for multi-view occlusion effects and is able to near-optimally fuse information from an arbitrary number of images collected opportunistically.

A version of the BMVS software has been optimized for commercial satellite image processing, and is capable of effectively using individual images from multiple orbits (cross-track multi-stereo), as well as in-track stereo images. A user-friendly GUI enables convenient selection of data and configuration of BMVS for processing. Quantitative algorithm performance data and algorithm performance comparisons are available upon request. An example reconstructed digital surface model (DSM) is shown in the figure below, along with “ground truth” data of the same scene. The x- and y-axes represent East and North, respectively, while the pixel intensities represent 3D altitudes, with lighter intensities indicating higher altitudes (the color scales on the right specify altitude in meters).

Left: “ground truth” DSM (from LIDAR).

Right: reconstructed DSM (reconstructed using a single pair of satellite images).