With the advancement of aerial imaging sensors, high quality data equipped with partial sensor calibration models is available. There is a recent research activity in computer vision community that aims to reconstruct 3-d structure of the observed scenes relying on the content of the imagery in fully automated ways. However the research has not matured into robust systems ready for operational settings. In this proposal, a novel architecture that reconstructs the 3-d geometry of the scene in the form of a geo-registered 3-d point cloud given imagery from multiple sensor platforms is presented. The 3-d cloud is equipped with LE and CE measurements through propagation of errors in the sensor calibration and the geometry reconstruction stages. The CVG team proposes to use a volumetric probabilistic 3-d representation (P3DM) and dense image matching to reconstruct the geometry and the appearance of the scene starting from a set of images with partial calibration data. The P3DM technology is at Technical Readiness Level (TRL) 4, with critical modules of the system parallelized and implemented on GPU hardware for real-time processing.
Keywords: Geo-Registration, 3-D Point Cloud, Camera Calibration, Unmanned Air Vehicle, Uav, Uas, Aerial Imagery, 3-D Reconstruction
