Real-time generic target tracking for structural displacement monitoring under environmental uncertainties via deep learning

Jong Hyun Jeong, Hongki Jo

Research output: Contribution to journalArticlepeer-review

Abstract

While structural displacement provides essential information about static and/or low-frequency dynamic characteristics of structural behaviors, full-scale measurement of absolute displacement in field structures is extremely challenging because of the requirement of fixed reference in most cases. Recent computer vision-based sensing technologies have advanced to the level of reference-free monitoring of full-scale dynamic displacement using generic features of the structure. However, current generic feature-based methods have limited to only short-term or campaign-type monitoring applications due to the intrinsic limitations of computer-vision sensing under variable environmental conditions. This study investigates deep learning-based approaches for real-time computer-vision sensing that enables displacement monitoring using generic features under harsh environmental uncertainties. Distractor-Aware Siamese Region Proposal Network (DaSiamRPN) was employed to address the environmental uncertainty issues, particularly caused by luminous condition change and obstructed vision, without sacrificing real-time processing capability. A series of indoor and outdoor experiments have been conducted to evaluate the performance under light condition change, occlusion, and haze. Comparative tests showed that the proposed method outperformed other various vision-based object tracking methods, showing the feasibility for long-term structural displacement monitoring of full-scale structures.

Original languageEnglish (US)
Article numbere2902
JournalStructural Control and Health Monitoring
Volume29
Issue number3
DOIs
StatePublished - Mar 2022

ASJC Scopus subject areas

  • Civil and Structural Engineering
  • Building and Construction
  • Mechanics of Materials

Fingerprint

Dive into the research topics of 'Real-time generic target tracking for structural displacement monitoring under environmental uncertainties via deep learning'. Together they form a unique fingerprint.

Cite this