In order to reconstruct the 3D shape of the non-structural surface of the hull plate frame accurately and quickly, and lay a foundation for the study on the high-precision and efficient measurement of the hull structure deformation, a model reconstruction method based on RGB-D depth image was designed.

First, the Random Sampling Consistency algorithm and the Least Square Method are combined to remove outliers in the point cloud, and then the checkerboard target position information of RGB color image is used to register the Multi-View Cloud of structure. Second, the point cloud is clustered by the regional grid of the structure surface, and the spatial curved surface of each grid point cloud subset is fitted by using the Least Square Method to realize point cloud fusion. On this basis, the high-order panel element is used to realize the 3D reconstruction of the hull structure outer plate surface. Finally, the accuracy of the model reconstruction method is verified by comparing the specimen reconstruction model with the laser scanning point cloud.

The results show that compared with the laser scanning point cloud, the root mean square error of the random points on the 3D reconstruction model of the test object is 1.02 mm. The modeling accuracy meets the needs of ship construction engineering, but the data acquisition time of the structure RGB-D depth image is negligible compared with the laser scanning.

The research shows that the 3D reconstruction of the non-structural surface of the hull plate frame can be accurately and efficiently completed based on the proposed method, which provides a strong data support for the deformation measurement of the hull structure.