MLS Point Clouds Compression Method Based On Object Category

Due to its all-weather and non-contact characteristics,LiDAR technology provides a reliable data source for real-time three-dimensional spatial data acquisition.The MLS point cloud data can continuously and quickly acquire high-precision,high-density three-dimensional coordinates of urban areas and roads,and has gradually become an important data acquisition method for urban planning,urban three-dimensional reconstruction,and other work.Therefore,the three-dimensional target detection,extraction and classification work based on vehicle point cloud data has also received attention from scholars in many fields such as surveying and mapping,computer graphics,and has achieved remarkable results,providing strong technical support for subsequent point cloud applications.However,the data acquisition is convenient and at the same time,the problem is a large amount of point cloud data.After the target detection,the point density obtained by the classification data is often much larger than the actual needs,and there is a lot of redundancy.Therefore,this paper focuses on analyzing the data characteristics of the artificial ground point cloud and the tree point cloud in the vehicle scene,and on this basis,the compression of the artificial ground and the tree point cloud is studied.The specific research content is as follows:(1)For point cloud data with more regular shape man-made features,such as buildings,vehicles,roads,street lamps,etc.,a point cloud compression method for man-made objects that takes into account the geometric shape and density distributionis proposed.Methods The factors of displacement between the original point and the predicted point before and after Gaussian smoothing were analyzed.The weighted sum of the flatness change component and the density change component was used instead of the traditional curvature,normal vector and other surface variation metrics to rank the importance of points.The overestimation of point importance due to the scanning direction and the orientation of the object is suppressed,and at the same time,the influence of the deletion point on the local point set is considered in the iterative compression process.Through experimental verification,the artificial ground compression method proposed in this paper can adapt to a variety of artificial targets.(2)For point cloud with irregular surface,a tree point cloud compression method based on ambient light shielding is proposed.Firstly,the scanning direction of the scanner corresponding to the tree point cloud is found through the PCA algorithm,and then the ball in the direction of the ambient light generated by the scanning direction and the road data is divided into two to obtain the directional hemisphere corresponding to the trees.Finally,along the direction of the light,the shaded tree cloud is calculated and the visibility of each point in the point cloud is obtained.The visibility is used to sort the points,and the point cloud is compressed by deleting the lower point of importance.(3)Combining with the characteristics of vehicle point cloud data,from the point of view of application of point cloud data after detection,extraction,and classification,the point cloud data after compression of artificial objects,such as buildings and vehicles,and point clouds after tree compression are evaluated.In the evaluation of the compression results of artificial objects,the quality of the compressed point cloud data was evaluated based on the distance from the original point cloud to the compressed point cloud patches.In terms of evaluation of tree point cloud compression results,the tree parameters required for subsequent tree point cloud three-dimensional modeling,forestry applications,etc.were taken into account,and the tree point cloud compression quality was evaluated by comparing the tree parameter changes before and after compression.