A generic computer vision-based monocular six-degree-of-freedom displacement measurement method

Abstract

Monitoring six-degree-of-freedom (6-DOF) structural displacement via nondestructive measurement methods can help engineers assess structural safety. In this work, we propose a Unet3+ based six-degree-of-freedom Structural Displacement Measurement method (USSDM). It uses a UNet3+ fully convolutional deep learning network to extract targets from recorded videos. A corresponding displacement calculation method is proposed to replace conventional scale factor-based methods. The effects of the target edge width, edge length, brightness, and structural movement frequency on the accuracy and computational efficiency of the USSDM are experimentally examined. In this work, USSDM has average root-mean-square errors of 0.106 mm and 0.115${}^{\circ }$ in translational and rotational displacement measurements, respectively, and it has an FPS of 20.7. The USSDM can implement real-time and accurate 6-DOF displacement measurements. Its performance is compared with that of several existing methods. The attained models and codes are open for other researchers. Moreover, its potential applications and current limitations are discussed.