Robust In-Plane Structures Oscillation Monitoring by Terrestrial Photogrammetry.

Omar El-Kadi,Khaled Nassar,Adel El-Shazly

doi:10.3390/s20082223

Omar El-Kadi, Khaled Nassar + Show 1 more

Open Access

https://doi.org/10.3390/s20082223

Copy DOI

Abstract

Oscillation monitoring commonly requires complex setups integrating various types of sensors associated with intensive computations to achieve an adequate rate of observations and accuracy. This research presents a simple, cost-effective approach that allows two-dimensional oscillation monitoring by terrestrial photogrammetry using non-metric cameras. Tedious camera calibration procedures are eliminated by using a grid target that allows geometric correction to be performed to the frame’s region of interest at which oscillations are monitored. Region-based convolutional neural networks (Faster R-CNN) techniques are adopted to minimize the light exposure limitations, commonly constraining applications of terrestrial photogrammetry. The proposed monitoring procedure is tested at outdoor conditions to check its reliability and accuracy and examining the effect of using Faster R-CNN on monitoring results. The proposed artificial intelligence (AI) aided oscillation monitoring allowed sub-millimeter accuracy monitoring with observation rates up to 60 frames per second and gained the benefit of high optical zoom offered by market available bridge cameras to monitor oscillation of targets 100 m apart with high accuracy.

Highlights

Deformation monitoring is an essential task in the field of geomatics, with vast fields on applications as landside deformations monitoring, monitoring of slopes and rock stability, structures and bridges deformation monitoring, and many more
Different systems of monitoring using close-range photogrammetry were tested [4], some systems are adopted for commercial use as dynamic monitoring station system, which was commercialized by university of Bristol, at the United Kingdom, in 2003 [5], the use of off-shelf modern digital cameras became a concern, and was tested for monitoring application by many researchers achieving sub-millimeter precision for both static and dynamic deformations [6,7]
As convolutional neural networks (CNNs) uses a huge number of regions in an input image, resulting in the need for extensive computing powers and limiting the application of CNN on large images, despite consolidating the network layers by max-pooling operations

Summary

Introduction

Deformation monitoring is an essential task in the field of geomatics, with vast fields on applications as landside deformations monitoring, monitoring of slopes and rock stability, structures and bridges deformation monitoring, and many more. Vehicles (UMAV) used in aerial photogrammetry is often defined as a drone, the resolution of the drone’s detection depends on the altitude and the characteristics of the camera [1]. Modern quadcopters can achieve a spatial resolution of one to three centimeters [2], which can be used for landslide deformation monitoring [3]. Different systems of monitoring using close-range photogrammetry were tested [4], some systems are adopted for commercial use as dynamic monitoring station system, which was commercialized by university of Bristol, at the United Kingdom, in 2003 [5], the use of off-shelf modern digital cameras became a concern, and was tested for monitoring application by many researchers achieving sub-millimeter precision for both static and dynamic deformations [6,7].

Results

Discussion

Conclusion