Accurate Measurement of Segment Dislocation for Shield Tunnels Based on Binocular Vision Technology

Segment dislocation is a common and problematic issue for shield tunnel linings, which might significantly affect the structural safety and serviceability in severe cases. To improve the accessibility of segment dislocation data, binocular vision technology combined with deep learning algorithms is introduced in this article to realize accurate measurement of segment dislocation. Intrinsic and extrinsic parameters of a consumer-level binocular camera are calibrated first, and the accurate disparity for binocular segment images can be calculated through a deep learning stereo-matching algorithm. Skeletons of segment joints are then extracted, and the least-squares method is applied for line fitting. As a result, pixel coordinates of key points near the segment joints are obtained, and the three-dimensional camera coordinates of key points are calculated according to the triangulation principle. Finally, the segment coordinate system is established by key points, and a camera pose correction based on the transformation relationship between the two coordinate systems is proposed to calculate the segment dislocation accurately. Compared with the measurement results by weld gauges, the calculation results from the binocular vision after camera pose correction show satisfactory consistency. The measurement errors across different shooting distances (ranging from 0.3 to 0.6 m) are less than 0.7 mm on average. In the scenarios of left-tilted or right-tilted shooting, when the roll and pitch of the camera pose are within the range of 180°±15° and 0°±20°, respectively, the proposed approach also demonstrates strong robustness and yields accurate measurement results. This article provides a new paradigm and method for the accurate measurement of segment dislocation.