Research On Intelligent Recognition Of Block Erection Surface In 3D Scanning Point Cloud Of Block

Accuracy detection of block erection surface is an important part of assembling and erection process.The accuracy detection of block erection surface has a profound impact on the shipyard's reduction of rework,accelerated construction cycle and cost reduction during the construction process.In terms of the accuracy detection of block erection surface,the 3D scanner has certain advantages compared with the total station because of its characteristics of rapidity,non-contact,high precision,etc.,and will be more and more applied to accuracy detection of block.However,during the scanning process,the 3D scanner also records the surrounding blocks,the environment and other unrelated points in the point cloud data.This has caused some difficulties for the researchers to compare the theoretical values of block erection surface with point cloud.Therefore,from the perspective of help 3D scanner do accuracy detection of block erection surface,the point cloud of block erection surface is identified.Considering that the data obtained by using the 3D scanner for block erection surface has less training amount and more noise,this paper uses tribon software to model the block and export the dxf file,and then use the algorithm of randomly generating points in the triangle area to generate point cloud data for the block.Using this method,more block point cloud data can be obtained,and the quality of the data is higher.Secondly,this paper studies the labeling problem of point cloud data of block erection surface,and finds that the method of block point cloud data by using multiple segmentation planes can adapt well to the special spatial data distribution of block erection surface,combined with Python programming language and computer graphics to complete the block point cloud visualization and labeling software.Using this software,the efficiency of point cloud labeling is greatly improved,the point cloud labeling work of block erection surface is completed,and the data enhancement algorithm is used to expand the limited data.Considering that point cloud recognition based on multi-view images and voxel-based point cloud recognition method lose the original information of point cloud data,this paper adopts point cloud data-based recognition algorithm PointSIFT.PointSIFT's random center point sampling algorithm is less effective due to the special spatial distribution of point cloud data on block erection surface.This paper uses random sampling of the entire block and adds eight inputs for each point in the block point cloud.Features to enable the network to learnlocation information.The identification of block erection surface by the final network model obtained precision of 76% on the validation set and recall rate of 92%.At the same time,using the model to identify the block point cloud actually scanned by the 3D scanner,the point cloud of block erection surface can be predicted,which proves the effectiveness of the algorithm.