In recent years, the extraction of structural dynamic characteristics by using phase-based video processing has attracted considerable attention. Edge detection routine is oftentimes employed to obtain the quantified operational deflection shape (ODS) vectors of structures. However, this approach is unreliable because of intensive human supervision and correction. To reduce operational uncertainty, a hybrid computer-vision-based approach called edge detection-region labeling, which involves vision sensor preparation, bottom-hat filtering, edge detection, hole filling, and region labeling, was presented in this work to extract the quantified ODS’s. The performance of this method was firstly evaluated by conducting a lab-scale cantilever beam test and subsequently the phase-based video processing was applied to extract the dynamic characteristics of an arch dam model. The operational modal analysis (OMA) test was conducted on the benchmark dam model. In-plane motions of the dam were captured and processed to identify the natural frequencies of the dam. The structural ODS’s were quantified using the proposed method. A comparison of the modal parameters of the dam identified from the video data with those obtained in the OMA test revealed that the two sets of results were consistent, and the video processing approach was able to bypass the requirement of human supervision, which facilitates the application of the phase-based video processing for complex structures.
Read full abstract