The utilization of high-resolution 3D point cloud data is becoming more common to a variety of DoD applications and many photogrammetric data sets are collected via small Unmanned Aircraft Systems (sUAS). Although these data are common, they do not always contain information regarding their geolocation (horizontal or vertical) uncertainties. This work addresses a critical challenge when collecting sUAS-derived point cloud data by developing a methodology to predict per-point geolocation uncertainties without the need for extensive validation. This point is particularly relevant when considering data collections for operations when collection and analysis time needs to be minimized.  To accomplish this task, a full error budget accounting for a sUAS flight was conducted. This included calculating the DGPS errors and uncertainties on the ground control points and determination of structural errors produced in the 3D reconstruction software. Specifically, this work compared geopositional errors between using pseudorange code GPS (i.e., relying solely on the sUAS GPS) vs differential GPS (DGPS), as well as developed a model of structural errors produced in the Structure from Motion software. Here, the impact of utilizing ground control points during the 3D reconstruction process was tested against data when ground control was not used. It was found that the utilization of DGPS surveyed ground control points during the 3D reconstruction process reduces both relative and absolute errors of the 3D data by more than an order of magnitude. Next, two machine learning models were developed to estimate the per-point measurement errors caused by poor 3D reconstruction from the Structure from Motion software that approximates the true measured error. The result is a methodology to compute a comprehensive per-point error product for sUAS data which in a simulation or training environment adds an increased level of realism for the training mission.

Keywords
3D;GEOSPATIAL DATA

Additional Keywords