- 學位論文
結合三點透視相機姿態估算法與光流法於結構振動量測
Image-based structural vibration measurement using perspective-three-point pose estimation and optical flow methods
若您是本文的作者，可授權文章由華藝線上圖書館中協助推廣。
摘要
傳統影像量測結構振動最大的限制在相機必須在量測過程中維持不動，本文突破此限制，提出基於影像移動式量測結構響應的方法，以標定不動的三維點位反算出每幀影像拍攝時的相機姿態，達到移動式晃動補償的目的，為移動式量測結構振動的問題中提出一個全新的方向。 本文提出的演算法共分成3塊模組，首先使用光流法獲取結構上移動特徵點的振動訊號，由於移動式相機所引起的漂移訊號是無法被忽略的，所以第二塊模組使用三點透視法估算出每幀影像拍攝時的相機姿態，最後補償模組用來消除與優化晃動補償的效果。為了驗證相機姿態估算結果的正確性，本研究產生不同相機姿態下拍攝的虛擬影像資料集與實際精密機械線性移動的影像，驗證結果皆可以有效的估算出相機的姿態。 實際移動式相機量測過程中，面外移動是無法避免的。當相機面外移動時，比例因子會隨著相機與物體間的深度而產生變化。透過本研究開發的相機姿態估算程式獲得每幀影像相機的位置，可以解決比例因子隨相機深度變化的問題。結合本文提出的更新比例因子法與移動式補償演算法，成功消除移動式相機在面內與面外量測結構振動時的漂移，此外，補償後的訊號與加速規在頻域中具有相當高的一致性。最後，本文進行移動式相機量測結構損傷定位實驗，透過頻域分解法萃取結構在健康與損傷時的模態，再使用模態曲率法有效地定位出結構損傷位置。
並列摘要
Traditional image-based vibration measurement is constrained by the need for the camera to remain still during the process. However, this research introduces a novel method that overcomes this limitation by utilizing a moving camera for vibration measurement. The proposed approach involves calibrating stationary 3D points to determine the camera pose in each frame, thereby compensating for motion-induced from camera. Proposing a new perspective to tackle the problem of structure vibration measurement from moving camera. The algorithm comprises three modules. Firstly, the optical flow method computes the vibration signal from moving feature points on the structure. In the second module, affected by camera drift, the perspective three-point method is utilized to estimate the camera pose for each frame. Finally, the compensation module performs motion compensation and optimization. To validate the accuracy of camera pose estimation, virtual image datasets with varying camera poses and actual images from precise mechanical linear motion are generated. During practical movable camera measurements, out-of-plane motion is inevitable. Consequently, when the camera experiences such motion, the scale factor varies with the depth between the camera and the object. By utilizing the developed camera pose estimation program, the issue of scale factor variation due to camera motion in depth can be resolved. By combining the proposed motion compensation algorithm and the updated scale factor method, drifting signals can be effectively mitigated during both in-plane and out-of-plane measurements performed by the movable camera. Moreover, the compensated signal exhibits strong consistency with the accelerometer in the frequency domain. Lastly, the paper conducts experiments to locate structural damage using a movable camera. The proposed algorithm compensates for drifting and extracts the mode shape from intact and damaged states by using frequency domain decomposition. The modal curvature method is then employed to precisely locate the structural damage.
參考文獻
[1] J.-S. Chou and W.-T. Tu, "Failure analysis and risk management of a collapsed large wind turbine tower," Engineering Failure Analysis, vol. 18, no. 1, pp. 295-313, 2011.
[2] L. Luo and M. Q. Feng, "Edge‐enhanced matching for gradient‐based computer vision displacement measurement," Computer‐Aided Civil and Infrastructure Engineering, vol. 33, no. 12, pp. 1019-1040, 2018.
[3] C.-Z. Dong and F. N. Catbas, "A review of computer vision–based structural health monitoring at local and global levels," Structural Health Monitoring, vol. 20, no. 2, pp. 692-743, 2021.
[4] C. Harris and M. Stephens, "A combined corner and edge detector," in Alvey vision conference, 1988, vol. 15, pp. 10-5244.
[5] D. G. Lowe, "Distinctive image features from scale-invariant keypoints," International journal of computer vision, vol. 60, no. 2, pp. 91-110, 2004.