3D Urban Reconstruction from Wide Area Aerial Surveillance Video

We propose an approach to solve camera pose estimation and dense reconstruction from Wide Area Aerial Surveillance (WAAS) videos captured by an airborne platform hovering around the urban scenes. Our approach solves them in an online fashion: it incrementally updates a sparse 3D model as well as a dense 2.5D Digital Surface Model (DSM) as each new frame arrives; the camera pose of each new frame is estimated using Perspective-n-Point (PnP) method with 2D-3D image-model feature matches. Dense optical flow between successive frames computed after a step of 2-D stabilization is used to guide the feature matching between each new frame and the maintained sparse 3D model. Our approach provides an online solution for camera pose estimation and dense reconstruction, and is significantly faster than the latest batch methods. The camera poses are estimated as accurately as with global Bundle Adjustment without drift along the path. We also produce a highly-detailed full 3D model via volumetric integration. Experiments on both synthetic and real-world datasets validate its performance.