Spherical reticulated shell structure is an important structural form of large-span space buildings. It is of great significance to monitor three-dimensional (3D) dynamic responses of spherical reticulated shell structure to better understand its seismic performances, which will be helpful in the future to ensure the healthy condition of large-span space buildings during their lifespan. In this study, with the advantages of non-contact and high accuracy, a high-speed videogrammetric measurement method is proposed for monitoring the 3D dynamic responses of the seismically isolated, spherical, reticulated shell structural model. Two issues—the high-speed videogrammetric acquisition system and network configuration, as well as image sequence target tracking and positioning—are emphasized to achieve a cache of high-speed images and to improve the accuracy of tracking and positioning target points. The experimental results on the shaking table from the proposed method have been compared with those from traditional Optotrak Certus and accelerometers. The results prove that the proposed method is capable and useful for analyzing the seismic performance of spherical reticulated shell structures, as the dynamic responses monitoring accuracy of the method can reach the submillimeter level, with root mean square error values of 0.32 mm, 0.7 mm and 0.06 mm in the X, Y and Z directions, respectively.

High-speed videogrammetric measurements are widely used on shaking tables. However, during progressive collapse experiments, the protective string used to ensure the safety of personnel and the shaking table, mandated by safety considerations, can partially occlude the structural model. To address the problem of inaccurate tracking of the ellipse targets in image sequences due to the partial occlusion, this paper proposes a novel mutually guided tracking method for the partial occlusion situations. Firstly, the strategy of loopback detection is proposed to eliminate the cumulative errors with the initial tracking model and to replace the initial results of the tracking with those from the loopback detection. Secondly, tiny offset compensation is used to solve the problem of deviations. The experimental results demonstrate that the proposed method can achieve single-point localization at the sub-millimeter level and interlayer localization at the millimeter level within partially occluded environments. It is important that the proposed method meets the requirements of experimental accuracy on shaking tables and ensures the safety of personnel and facilities.