基于深度相机的古建筑木结构裂缝识别联合多分辨率三维重建方法

Crack Recognition and Multi-resolution 3D Reconstruction of Ancient Wooden Structures Based on Depth Camera

目的研究木结构古建筑的三维数字化技术以及木结构表面裂缝分析方法,为该类古建筑的裂缝测量和维护提供参考。方法通过Mask R-CNN对RGBD图像中的裂缝信息进行识别,结合位姿优化策略计算每一帧的相机姿态,进而获得裂缝的三维信息;构造一种包含古建筑三维模型与裂缝信息的多分辨率三维重建模型,基于所构造的模型分析古建筑表面裂缝的长度和深度。结果提高裂缝三维重建精度前,长度平均计算误差为12.3%,深度平均计算误差为36.8%;提高裂缝三维重建精度后,长度平均计算误差为9.3%,深度平均计算误差为19.7%。结论提出的识别方法可以快速、精准地识别古建筑木结构中的裂缝并获得古建筑的三维模型,相较于传统裂缝测量效率更高、成本更低。

This paper was proposed to study the 3D digital technology and surface crack analysis method of wooden structure ancient buildings,and to provide reference for the modeling and crack measurement of this kind of ancient buildings.On the basis of Mask R-CNN to identify the crack information in RGBD images,the local to global optimization strategy was used to calculate the camera pose,and a multi-resolution implicit structure to contain the 3D model of ancient architecture and crack information and analyze the length and depth of cracks on the surface of ancient buildings.A multi-resolution 3D model with fracture information was obtained,which improved the accuracy of 3D reconstruction of the fracture,and analyzed the length and depth of the fracture.In Red Wood data set and real scenes,the average calculation error of fracture length and depth is 12.3%and 36.8%respectively before improving the 3D reconstruction accuracy.After improving the 3D reconstruction accuracy,the average calculation error of length and depth is 9.3%and 19.7%respectively.It is proved that the method can quickly and accurately identify cracks in ancient wooden structures and obtain 3D models of ancient buildings,which is more efficient and cheaper than traditional crack measurement.