Achieving Tiered Model Quality in 3D Structure from Motion Models Using a Multi-Scale View-Planning Algorithm for Automated Targeted Inspection

Abstract

This study presents a novel multi-scale view-planning algorithm for automated targeted inspection using unmanned aircraft systems (UAS). In industrial inspection, it is important to collect the most relevant data to keep processing demands, both human and computational, to a minimum. This study investigates the viability of automated targeted multi-scale image acquisition for Structure from Motion (SfM)-based infrastructure modeling. A traditional view-planning approach for SfM is extended to a multi-scale approach, planning for targeted regions of high, medium, and low priority. The unmanned aerial vehicle (UAV) can traverse the entire aerial space and facilitates collection of an optimized set of views, both close to and far away from areas of interest. The test case for field validation is the Tibble Fork Dam in Utah. Using the targeted multi-scale flight planning, a UAV automatically flies a tiered inspection using less than 25% of the number of photos needed to model the entire dam at high-priority level. This results in approximately 75% reduced flight time and model processing load, while still maintaining high model accuracy where needed. Models display stepped improvement in visual clarity and SfM reconstruction integrity by priority level, with the higher priority regions more accurately modeling smaller and finer features. A resolution map of the final tiered model is included. While this study focuses on multi-scale view planning for optical sensors, the methods potentially extend to other remote sensors, such as aerial LiDAR.

Geodetic and UAV Monitoring in the Sustainable Management of Shallow Landslides and Erosion of a Susceptible Urban Environment

“…Throughout complex and precise parameters, processing power, and software, based on projective geometry it converts the data extracted from the images into three-dimensional metric coordinates and colors. Given the present case study, difficult terrain conditions of the monitored slope associated with precise land surveying, this paper assesses the traditional monitoring of established techniques and instrumentation with the use of UAV that ensures accuracy and viability as an affordable and efficient method that accompanies traditional surveys for more comprehensive results [54][55][56][57][58][59].…”

Feasibility Assessments Using Unmanned Aerial Vehicle Technology in Heritage Buildings: Rehabilitation-Restoration, Spatial Analysis and Tourism Potential Analysis

“…The framework of 3D model consisted of similar steps, starting with images acquired manually, followed by a pre-selection to remove blurred photos, image rectification, point cloud segmentation and 3D reconstruction ( Figure 6). Taking into account the laws of multiple-view geometry, images of the desired model should be collected from all possible viewing directions, in order to reconstruct its accurate 3D shape and requires a camera network that possess nadir and oblique images with various baselines [23][24][25][26]. Combining the mentioned steps, adequate obtained data and the specialized software with processing power, it is viable to obtain a 3D point cloud of the heritage building and environment within a few centimeters accuracy (Figure 7).…”

Automated 3D Reconstruction Using Optimized View-Planning Algorithms for Iterative Development of Structure-from-Motion Models

“…In these studies 3D point clouds serve as DEMs and assist in eventual model analysis. Another example, though not truly a complex 3D object nor an overhang-is a multi-scale model of the Tibble Fork Dam in Utah County, UT, made possible by the DEM of the scene whose sub-sampled point cloud lends the confidence to safely fly the UAV on successive flights closer to the dam, model quality at key points of interest and maintain the integrity of the entire model without inordinate amounts of data input or time [15].…”

“…An example application of this strategy would be to use the iterative approach to map an unknown area that is inaccessible to humans due to hazardous conditions but where the GPS coordinates are known. After an initial model has been obtained , it is recommended to use refinement tools, such as the work presented in Okeson et al That provides targeted reconstruction to desired levels of accuracy [15].…”

“…The success of SfM and UAVs heightens the interest in developing robust frameworks for industrial monitoring and inspection. Researchers have addressed tasks such as performance in unknown environments [8], computational requirements [9], autonomy [10], plant breeding [11], deep learning in agriculture [12], multi-spectral imaging [13], and view planning [14,15]. This study introduces an iterative modeling strategy that provides UAVs with additional autonomy in photogrammetry missions.…”