Full-field structural vibration monitoring and system identification using deep learning and optical flow

摘要：

　　The prevalent vision-based displacement measurement methods such as digital image correlation and feature point-based method mostly measure the displacement of a single point or several points,not the full field measurement which means to obtain the displacement at each pixel in the field of view.Although therere some algorithms of optical flow which can achieve full field measurement,they still need lots of human manual interventions in implementation to adjust parameters such as thresholds and window size.Also,they rarely leverage the spatial and temporal context in the image during tracking and without learning process in image sequence.This study tries to develop a new vision-based displacement measurement method,which can eliminate the problems that the prevalent methods are facing and achieve full field displacement measurement with less human manual intervention,no manual marker,taking ad-vantage of spatial and temporal context in the image sequence.Eventually,an adaptive monitoring and end-to-end learning framework is developed.The feasibility of the proposed method is validated with a series of experiments on a grand-stand in laboratory,comparing with the conventional displacement sensor and a prevalent visionbased displacement measurement method.

关键词： Structural health monitoring dynamic displacement measurement optical flow deep learning

作者: Chuan-Zhi Dong Ozan Celik F.Necati Catbas

作者单位: Department of Civil,Environmental,and Construction Engineering,University of Central Florida,Orlando,FL 32816,USA

会议类型: 国际会议

会议名称: The 7th World Conference on Structural Control and Monitoring(7WCSCM)(第七届结构控制与监测世界大会)

会议地点: 青岛

会议语种:英文

页码: 1560-1567

在线出版日期: 2018-07-22（万方平台首次上网日期，不代表论文的发表时间）