Computer vision-based illumination-robust and multi-point simultaneous structural displacement measuring method

•Developed a computer vision-based method for structural displacement measurement.•Employed fully convolutional network (FCN), conditional random field (CRF) and digital image correlation (DIC) related techniques.•Performed validation experiments on a simply supported beam.•Verified illumination-robust and multi-point simultaneous measurements. Computer vision-based techniques for structural displacement measurement are rapidly becoming popular in civil structural engineering. However, most existing computer vision-based displacement measurement methods require man-made targets for object matching or tracking, besides usually the measurement accuracies are seriously sensitive to the ambient illumination variations. A computer vision-based illumination robust and multi-point simultaneous measuring method is proposed for structural displacement measurements. The method consists of two part, one is for segmenting the beam body from its background, the segmentation is perfectly carried out by fully convolutional network (FCN) and conditional random field (CRF); another is digital image correlation (DIC)-based displacement measurement. A simply supported beam is built in laboratory. The accuracy and illumination robustness are verified through three groups of elaborately designed experiments. Due to the exploitation of FCN and CRF for pixel-wise segmentation, numbers of locations along with the segmented beam body can be chosen and measured simultaneously. It is verified that the method is illumination robust since the displacement measurements are with the smallest fluctuations to the illumination variations. The proposed method does not require any man-made targets attached on the structure, but because of the exploitation of DIC in displacement measurement, the regions centered on the measuring points need to have texture feature.