Estimating structural motions in extreme environmental conditions--A dynamic correlation filter based computer vision approach

Vision-based methods have shown great potential in vibration-based structural health monitoring (SHM). However, these methods are not standard practices yet, since their accuracy and robustness may be influenced by extreme environmental conditions. To this end, this paper proposed a method, named dynamic regularized total variation correlation filter (DTVCF). In DTVCF, an effective optimization problem, which contains the space and structural shape information, is defined for dynamically updating the regularization weight map. Then the regularization weight map is smoothed by a total variation (TV) optimization. These are to better track the dramatically changing or almost invisible structural shape, caused by extreme environmental conditions. Moreover, efficient subpixel image registration (ESR) is used in each tracked region of interest (ROI), over time, to achieve subpixel accuracy. The superiority of DTVCF was validated in extreme environmental conditions. DTVCF could achieve subpixel level structural displacement estimation with high accuracy. Furthermore, DTVCF could process approximately 9.00 frame/s, and 3.50 frame/s in two shaking table tests, indicating its high efficiency for SHM applications.