Flexible Multicamera Virtual Focal Plane: A Light-Field Dynamic Homography Approach

Combining images from multiple cameras using computational imaging techniques is a common way in remote sensing to achieve a larger field-of-view (FOV) and wider scene coverage using manned and unmanned aerial vehicles (UAVs). This chapter proposes an analytical approach, called Light-Field Dynamic Homography (LDH), to create a large virtual focal plane from an array of nonrigid flexibly positioned cameras with narrow overlaps between their FOVs for remote sensing and Wide-Area Motion Imagery (WAMI). The proposed LDH method dynamically estimates the view-to-view transformations between a set of flexible cameras by taking their underlying 3D geometry into account using a nonlinear least squares optimization approach. We establish that the minimum number of point or feature correspondences between each adjacent pair of image views for LDH is just two. This relaxes the need for large spatial overlaps in the overlapping camera FOVs. Extensive simulation experiments characterize the influence of several parameters on LDH accuracy including the minimum number of point correspondences, the amount of overlap between adjacent views, and radial distortion on the precision and integrity of the virtual focal plane light field. Our simulation studies indicate the repeatability and superior performance of the proposed 3D-based LDH method compared to approaches using traditional 2D methods such as projective or affine homographies.