PSI - Issue 78

Laura Gioiella et al. / Procedia Structural Integrity 78 (2026) 1436–1442

1437

1. Introduction Shake table testing is a key experimental approach for evaluating the seismic performance of structures as reported in Van Den Einde et al. (2021) and Nakashima et al. (2018). Its success depends on the accurate reproduction of ground motions and on precise measurement of structural response parameters, among which absolute and relative displacements are critical. Conventional displacement measurements often rely on contact sensors, as shown in Pei et al. (2024) or on double integration of accelerometer data, Bock et al. (2011). Both methods face drawbacks: contact sensors require fixed reference points and time-consuming installation, while acceleration integration can introduce significant errors despite corrective processing, as noted by Skolnik and Wallace (2010) and Dai et al. (2020). Computer vision offers a non-contact alternative by extracting motion data from image sequences recorded during testing as documented in Feng and Feng (2021). Initially applied mainly to bridges, Zona (2021), its use in building monitoring has expanded, as demonstrated by the studies of Ngeljaratan and Moustafa (2020)–Zhou et al. (2024). However, common setups—external tripod-mounted cameras, Ngeljaratan and Moustafa (2020)–Sun et al. (2025); Unmanned Aerial Vehicle-mounted devices, Wang et al. (2022); or internal installations, Park et al. (2010)–Zhou et al. (2024),—present limitations related to coverage, synchronization, motion compensation, or lack of a stable reference frame. This work proposes a streamlined vision-based method for measuring horizontal displacements in multi-story buildings, uniquely combining an internal roof-mounted camera and an external ground-based camera. The roof camera captures multiple targets along the building’s height, while the external camera records the roof displacement near the internal camera, providing redundancy and reducing noise. Although developed for shake table applications, the approach is adaptable to other dynamic tests, such as push-and-release experiments proposed by Dall’Asta et al. (2022). Its effectiveness is demonstrated through the shake table testing of a six-story mass timber structure on the 6 DOF Large High Performance Outdoor Shake Table (LHPOST) at the University of California, San Diego (UCSD). 2. Methodology The proposed vision-based methodology can be deployed in two configurations: a planar setup with two cameras, or a three-dimensional setup with triplet cameras. In both cases, displacement measurements are obtained using the Upsampling Cross-Correlation (UCC) algorithm, Guizar et al. (2008), which achieves sub-pixel accuracy—up to about 1/100 of a pixel—when combined with high-contrast markers such as chessboard-pattern targets with varied patterns and sizes, as shown by Micozzi et al. (2023). 2.1. Planar configuration In its simplest form, the method uses two cameras to monitor a building subjected to a mono-directional shake table excitation (Fig. 1). An internal roof-mounted camera (light blue in Fig. 1) points downward to track multiple upward targets (T0–T4) placed along the building height to prevent overlap during motion. It also measures shake table displacement from the relative motion of T0 and T1, enabling comparison with the table controller. An external camera (light green in Fig. 1) is fixed on the ground-level far from the reaction mass (L0) and records the roof displacement via target T5, located close to the internal camera. Thus, the roof motion (L5) is measured twice: from T5 by the external camera and from T0 by the internal one. In ideal conditions these measurements match; differences reveal vibrations or rotations of the internal camera due to the shaking, producing spurious motion in T0–T4. Redundancy enables noise compensation by subtracting the displacement of T0 from T5, after which compensated floor displacements and inter-story drifts can be calculated.