Theory and development of a camera-based noncontact vibration measurement system

Abstract

Dramatic advancement in technologies for high-speed high-resolution digital cameras in recent years enables the development of camera-based full-field noncontact measurement systems for vibration testing of flexible multibody systems undergoing large rigid-body motion and elastic/plastic deformations. A few of such systems exist in today's metrology market, but they are inconvenient for use and prohibitively expensive. Most seriously, they are not really appropriate for structural vibration testing because their measurement accuracy is low due to several technical reasons, including inappropriate setting of cameras and experimental setup because of user's innocence of video-grammetry, non-precise corner detection and other problems of image processing techniques, and inaccurate modeling and calibration of cameras. This thesis develops and puts together a complete set of necessary techniques for the development of a camerabased noncontact full-field vibration measurement system using inexpensive off-the-shelf digital cameras. An optimal combination of appropriate methods for corner detection, camera calibration, lens distortion modeling, and measurement applications is proposed and numerically and experimentally verified. Moreover, we derive/improve some image processing methods and 3D reconstruction algorithms to improve vibration measurement accuracy. The proposed methods include: 1) a corner detection method for processing 2D images with sub-pixel resolutions, 2) an improved flexible camera calibration method for easy and fast calibration with high accuracy, 3) a lens distortion model for correcting radial, decentering, and thin prism distortions, 4) a set of guidelines for setting up cameras and experiments for measurement, and 5) algorithms for measurement applications. The proposed corner detection method improves Foerstner's corner detector, which improved Moravec's and Harris's corner detectors. The proposed camera calibration method improves Zhang's flexible technique, which works without knowing the object's 3D geometry or computer vision. The method only requires the camera to observe a planar pattern (e.g., a checker board) shown at two or more independent orientations by arbitrarily moving the planar pattern (or the camera). Estimation of the camera's intrinsic parameters (i.e., focal length, principal point, the skewness parameter and aspect ratios of the two image axes, and lens distortion parameters) and extrinsic parameters (i.e., camera's location and orientation with respect to the referential world coordinate system) consists of an approximate initial guess based on linear closed-form solutions and then nonlinear optimization for refinement. This approach is between the photogrammetric calibration and the self-calibration. Compared with photogrammetric calibration techniques that use expensive calibration objects of two or three orthogonal planes, the proposed technique is easy to use and flexible. To examine the proposed methods and their combined effects against high measurement accuracy, two Canon EOS-7D DSLR cameras are used for theoretical studies and experimental verifications. Numerical and experimental results show that the recommended methods together with our improved image processing techniques is feasible for the development of a camera-based noncontact full-field vibration measurement system with high precision and low cost. This camera-based measurement instrument has the potential for developing new structural testing techniques and can open new possibilities for research and development in mechanical and aerospace engineering, computer science, animal science, and many other fields.