High Resolution 3D Modeling Using Oblique Pictometry and Lidar Data

Abstract

As part of a larger project to develop a high resolution model of the Virginia Tech campus, we processed over 8,000 non-georeferenced aerial oblique images of Blacksburg area collected by Pictometry in 2019. We sequentially: (a) produced an initial camera position estimate from image footprints in Python, (b) calibrated the image set by creating approximately 200 ground control points (3D GCPs using position and elevation) and over 2,500 image marks manually generated with Google Earth, and (c) after adding final fine referencing using RTK GPS, we calculated the 3D original camera positions using Pix4D software. This challenging project used unconventional methods to establish camera location and orientation by using imagery that was not created with 3D modeling in mind (i.e. low image overlap) and calibrating model cameras using Google Earth derived data for GCP construction. Finally, we used RealityCapture software to fuse lidar imagery with our georeferenced image set to produce a 3D model that combines the spatial accuracy of lidar with the high point density of Structure from Motion (SfM) models. We expect to use the final constructed model for several applications, including to support indoor mapping and navigation and interactive, augmented reality 3D printed maps for people with visual impairments.