虚拟现实动态建模的设计与实现

Design and implementation of virtual reality dynamic modeling

针对服务于智能驾驶的虚拟现实动态建模,基于点云数据的四面体剖分进行优化研究。整个过程包括点云数据抽取、数据增强、多层卷积、多层感知、智能决策支持、数据桥接和动态建模。在分类识别阶段,针对激光雷达点云数据的稀疏性,通过Delaunay四面体剖分结合旋转和缩放进行数据增强,采用PointNet深度神经网络进行模型训练。测试结果表明经过数据增强后的分类准确度可以提高到90.6%。在动态建模阶段,在分类识别的基础上,通过CAD模型辅助建模,采用Delaunay四面体剖分建立立体模型,2个阶段通过数据库相耦合。动态建模结果验证了该方案的可行性。

Based on the tetrahedral subdivision of point cloud data,an optimization method for virtual reality dynamic modeling in intelligent driving was proposed.The complete process involved the extraction of point cloud data,data enhancement,multi-layer convolution,multi-layer perception,intelligent decision support,data bridging and dynamic modeling.In the classification recognition stage,due to the sparse LiDAR point cloud data,data enhancement was made through the combination of Delaunay tetrahedral subdivision,rotation and scaling.The model training was finished by PointNet deep neural network and the results of the tests indicated that the accuracy could be increased to 90.6%.During the dynamic modeling phase,CAD models were adopted based on the previous classification recognition to assist modeling and Delaunay tetrahedral subdivision was utilized to generate three-dimensional models.The two phases were coupled by database.The feasibility of the proposed method was verified through dynamic modeling results.