机器视觉运动放大及其在结构模态识别的应用进展
Recent advances of computer visionmotion amplification and its application in modal identification of structures
Received:March 16, 2021  Revised:June 11, 2021
DOI:10.7520/1001-4888-21-056
中文关键词:  运动放大  模态分析  机器视觉
英文关键词:motion magnification  modal identification  computer vision
基金项目:国家自然科学基金 (51778103);国家自然科学基金 (52078284);广东省自然科学基金 (20210001);汕头大学科研启动基金(NTF18012) 资助
Author NameAffiliation
LI Dongsheng* Guangdong Engineering Centerfor Structure Safety and Health Monitoring, Department of Civil and Environmental Engineering, Shantou University, Shantou 515063, Guangdong, China 
LIN Jianyuan Guangdong Engineering Centerfor Structure Safety and Health Monitoring, Department of Civil and Environmental Engineering, Shantou University, Shantou 515063, Guangdong, China 
CAI Enjian Tsinghua University, Department of Civil Engineering, Beijing 100084, China 
LI Hongnan State Key Laboratory of Coastal & Offshore Engineering, Dalian University of Technology, Dalian 116023, Liaoning, China 
Hits: 324
Download times: 0
中文摘要:
      机器视觉运动放大技术能将肉眼难以捕捉的细微结构振动响应进行放大,依据不同空域滤波器将原始图像分解成不同的子图像,从而获取传统方法难以识别的高阶模态信息。本文首先介绍了机器视觉运动放大的拉格朗日方法和欧拉方法,并从噪声处理以及大运动隔离两方面讨论了最新的进展;然后论述了数字图像相关技术、点追踪技术和相位光流法三种提取图像运动特征的方法;之后介绍了本文作者如何利用亮度方程进行相位展开修正原始复值可操控金字塔模型、移除2π相位阶跃避免产生伪像的方法及研究进展;PVC管道振动实验表明高放大倍率下运动放大技术可较好地提取结构高阶模态信息;最后,讨论了视觉运动放大方法在实际应用中存在的优缺点,提出相应的改良措施,并展望了其未来发展方向。
英文摘要:
      Motion amplification technology can amplify the subtle structural vibration response that is difficult to capture by eyes, decompose the original image into sub-sampling sequences at different spatial resolutions, and identify the high-order modal information that cannot be obtained by traditional technology. Firstly, the Lagrange method and Euler method of machine vision motion amplification are presented, moreover, some new methods of noise processing and large motion isolation methods are introduced respectively. Then, three methods of extracting the motion feature of image are summarized, such as digital image correlation technology, point tracking technology and phase optical flow method. In this paper, the author introduces the research progress of correcting the original complex value controllable pyramid model by using the intensity equation for phase unrolling and removing the 2π phase step to avoid artifacts. Examples of the PVC pipe show that the motion amplification technique can extract the high order mode information of structure well under high magnification. Finally, this paper mainly discusses the defects of the current machine vision motion magnification method in the practical applications, puts forward the corresponding improvement measurements, and predicts the direction of future development.
View Full Text  View/Add Comment  Download reader
Close