基于三维重建的船舶合拢管测量技术

Measurement Technology of Ship’s Inserting Pipe Based on 3D Reconstruction

文中将结合深度学习的三维重建技术引入合拢管测量中,实现法兰空间姿态的复现,利用深度学习网络能自动提取输入图像高全局特征的优势,改善多视角立体匹配的稳定性和精度,获得完整、稠密的三维重建点云.利用基于Open3D设计算法处理重建结果,获取船舶合拢管法兰之间的相对位姿.对两种不同位姿的合拢管法兰测量,测量的平均误差均在0.5 mm以内,测量时间均在3 min以内,其测量精度和效率可满足合拢管的制造需求.

Three-dimensional reconstruction technology combined with deep learning was introduced into closed pipe measurement to realize the reproduction of flange spatial posture.Using the advantage that the deep learning network can automatically extract the high global features of the input image,the stability and accuracy of multi-view stereo matching could be improved,so as to obtain a complete and dense three-dimensional reconstructed point cloud.The reconstruction results were processed by Open3D design algorithm,and the relative position and posture between the flanges of the ship’s closed pipe were obtained.The average error of flange measurement for two kinds of closed pipes with different postures is less than 0.5mm,and the measurement time is less than 3 minutes.The measurement accuracy and efficiency can meet the manufacturing requirements of closed pipes.