Research On Pavement Point Cloud Modeling Parallel Algorithm And Defect Detection Method

In recent years,the total length of roads at the management and maintenance stage in China has reached the second place in the world.In addition to the low efficiency and accuracy that relying solely on manual labor,there are also problems in that the total mileage is too large to be completely covered by human work.In order to conduct road maintenance more efficiently,a movement detection system made up of a test vehicle equipped with a high-precision laser sensor set has become a research hot spot.This system can automatically obtain the road surface point cloud and then process it to obtain a high-precision road surface model.However,due to the large scale of the point cloud collected by the system,the modeling relying on the traditional method runs too slowly.Therefore,it is necessary to parallelize various issues in the modeling of the road surface point cloud to increase the operating speed.This paper first studies the parallel acceleration of large-scale laser point cloud data preprocessing.Point cloud data preprocessing includes mend and simplification.The process of processing is to first patch the data of the missing location,and then the complemented point cloud is simplified to obtain a feature point cloud.The point cloud mend adopts the patching algorithm that assigns the missing points.The point cloud simplification uses the algorithm based on the slope difference of the scanning line point cloud.The road surface point cloud belongs to a type of scan line point cloud,so it has the feature that mends or simplifies each scan line does not affect other scan line points,so the paper proposes a parallel idea to use one thread to process one scan line.In this way,parallel algorithms based on multi-core CPUs and GPU parallel algorithms are implemented and compared.In this paper,we study the parallel problems of irregular triangulation modeling on the feature point cloud obtained by preprocessing.In order to meet the accuracy requirements,the model to be established is the Delaunay triangulation.However,most of the traditional Delaunay triangulation algorithms are based on three-dimensional point cloud datasets,and they are less efficient to perform on large-scale point cloud datasets.The point cloud data needs to be reduced dimensionally,and the 3D point cloud is converted into a 2D point cloud to reduce the computational burden.On this basis,a parallel triangulation algorithm based on region decomposition and convex hull consolidation is implemented,and the point cloud is divided by region decomposition to obtain a plurality of sub-point sets,then performs triangulation on all the sub-point sets to get a sub-triangulation,and finallymerges the boundary convex hulls of all the sub-triangulations to generate a complete triangulation model.The paper also studies the automatic detection of pavement defects.Based on the Delaunay triangulation model,the classification is based on the difference in the geometric characteristics between the triangles of the pavement defect area and the normal triangles,and also adopt the least square method fitting the road cross section to estimate the defect location.In order to evaluate the parallel efficiency of the above algorithms,a total of 20 km of actual road surface point clouds were collected as test data.The parallel implementation of the point cloud preprocessing and the parallel implementation of delaunay triangulation have achieved better acceleration effect.This method can be used for online pavement modeling below 80 km/h.Finally,a 3D road surface modeling system is designed and implemented with Open GL.The system can display different road surface models,and at the same time,it can freely switch different observation perspectives to roam the model.