A robust vision-based method for structural response measurement using UAV and cross-laser reference

Abstract

Structural responses, such as displacement and rotation, are crucial parameters that reflect structural performance, and thus their accurate measurement is essential for structural safety assessment. Computer vision techniques that enable response extraction from image sequences have gained significant attention. Most vision-based measurement methods using fixed cameras often require long imaging distances, which reduce image resolution and measurement accuracy. Unmanned aerial vehicles (UAVs) enable close-range image acquisition from optimal angles due to their high maneuverability, overcoming the spatial limitation of the fixed camera. Existing UAV-based methods generally require stationary reference features to correct UAV-induced errors. Sufficient stationary reference features are often unavailable for bridges crossing rivers or roads. Moreover, illumination variation, motion blur, and partial occlusion in practical measurements may further degrade feature extraction and measurement accuracy. This paper proposes a robust vision-based method for structural response measurement using UAV and cross-laser reference. By introducing a cross-laser projection as a stationary spatial reference, this approach enables structural response correction from the relative geometric relationship between the cross-laser and marker without the camera pose information. A robust localization method combining multi-frame fusion-assisted Otsu segmentation and geometric fitting is proposed to achieve sub-pixel localization of marker and cross-laser pattern. The cross-laser projection provides positional and directional reference, allowing for simultaneous displacement and rotation measurement by tracking the intersection point and orientation variations. The feasibility and robustness of the proposed method are verified through laboratory experiments on a truss bridge under various influencing conditions, where the displacement and rotation measured by the vision-based method agree with those measured by contact sensors.