To address the issue of large rebar dimensions that cannot be covered by a single camera field of view in processing and inspection,a method for measuring the geometric shape(length and angle)of rebar using computer vision technology was proposeed.A partitioned camera measurement system was proposed.The partitioned cameras were utilized to simultaneously capture images and stitch together a panoramic view based on binary encoded ArUco markers.For real-time computing requirements,a lightweight semantic segmentation network,named as ghost-Unet,was proposed to balance the efficiency and accuracy of rebar contour recognition.To ensure computational accuracy,a robust straight-line detection algorithm was utilized for segmenting and fitting the rebar contours.The measurement results indicate that in the laboratory,the average length measurement error is 2.92 mm when measuring vertical rebar with the optical axis of camera perpendicular to the plane.The average length measurement error is 4.27 mm when measuring horizontal rebar with the optical axis of camera at an angle much less than 90° to the plane due to a rebar thickening issue in the orthographic correction stage.In the field,the average length measurement error is 3.08 mm,and the angle error is within 0.7°.
关键词
施工质量检测/钢筋尺寸测量/图像拼接/计算机视觉/轻量化语义分割模型
Key words
construction quality inspection/measurement of steel bar size/image stitching/computer vision/lightweight semantic segmentation
维普
万方数据