Displacement measurement and nonlinear structural system identification: A vision-based approach with camera motion correction using planar structures

Knowledge of the camera motion is important in the applications of structural dynamic response measurement using a vision-based approach because in most of the field measurements, such motion can be non-trivial and the accuracy of dynamic response measurements is strongly affected by the camera motion. This paper presents a new framework for camera motion estimation and vision-based displacement measurement, which greatly lowers the barriers to the application of generally positioned cameras from strictly stationary cameras. The contributions in this paper are twofold. First, camera motion as well as reconstructed structural displacement are calculated based on reference planar structures visible in a given scene. In this case, a homography is more effective at describing view changes and planar geometric constraints can be incorporated early in the reconstruction process, thereby improving the quality and effectiveness of the estimates. The second contribution is that the homography of the planar structure is estimated by a newly proposed tracking algorithm that combines RANSAC algorithm and Efficient Second-order Minimization (ESM) technique, which refines the final estimates to sub-pixel accuracy and avoids tracking drift and non-smoothness effectively. Experimental results indicate that the quality of the camera motion estimation and displacement reconstruction can be significantly improved by the judgmatical use of the proposed algorithm for planar structure homography estimation. Furthermore, nonlinear structural system identification is carried out to additionally verify the proposed algorithm using unscented Kalman filter.