A Research On Realistic 3D Reconstruction Of Highway For Driving Simulation

With the road construction and the automobile industry technology development,road driving safety research has become the focus of attention of all countries.at present,Intelligent connected vehicle is the mainstream technology of automobile development.Limited by the complex driving environment,intelligent vehicle has not yet reached the level of autonomous driving.The driving simulation technology which combines virtual reality and cars is an efficient,safe and reliable training method.The driving simulator is limited by the influence of the driving traffic environment that cannot be fully replicated,and the fidelity is not high.The existing visual model of driving simulation is mainly based on the experience of the developer.It is produced by 3D dynamic simulation graphics generator and a large-screen high-definition display.The visual design seldom pays attention to the use of geographic information database(GIS),which has a certain gap with the real driving scene,especially road modeling.Based on the above background,a 3D digital highway model was put forward by utilizing heterogeneous multi-source data extracted from geographic information system.For implementation of the objective of our study,the steps were taken in sequence as follows:(1)2D roadway model reconstruction based on remote sensing imageThe remote sensing image is selected as the data set,and road extraction method is proposed by improved Seg Net network segmentation based on the combination of raster data and vector data.First,combined with the geometric and topographic features of the highway in the high-resolution remote sensing image,the road is regarded as an independent area,and the road vector data is used as the prior knowledge to mark the remote sensing image.The Seg Net network is used to segment the remote sensing image data by transfer learning.The road feature is extracted by the cavity convolution to realize the recognition of the highway image.Finally the image is post-processing and parameterization.Simulation experiments verify the reliability and versatility of the reconstruction algorithm.(2)Digital elevation model construction based on multi-source geographic information data fusionAiming at the accuracy and scale of the existing global digital elevation data sets,a high-precision data elevation model based on multi-source data fusion is proposed.First,spline interpolation is used to register the SRTM3 DEM data,which is consistent with the resolution of ASTER GDEM;then the problem of void outliers in the original data set is solved by elevation evaluation and weight estimation;secondly,The accuracy of the raw DEM data is improved through neural network,and finally the improved data set is weighted fused to generate a high-precision DEM.Three typical domestic geomorphological features are selected as data sets,it shows that the fusion results have higher results than the single DEM data and the interpolated DEM data.(3)Building 3D road visual model based on real geographical environmentThis paper explain the characteristics of raster data and vector data in GIS.The extracted 2D raster road data and the fused DEM are transformed into vector data through space vector transformation,which realizes the visualization of digital 3D model of expressway.With the help of World Machine terrain editing and Unreal Engine 4(UE4)software,the DEM raster data are imported to generate the road terrain model based on real geography.The road texture information extracted from remote sensing image is mapped to the road model surface to enhance the reality of the road surface.The 3D road data are imported into UE4 to restore the real road,and the 3D road view model based on specific geography is established.The road database is established to meet the needs of the virtual driving environment.(4)The realization of driving simulation in realistic 3D ReconstructionsThe evaluation is based on the effectiveness of the real road visual model.Under the road visual model,a 3D vehicle model is established in UE4 software to simulate the vehicle’s operating state in a typical highway scene.The verification is based on the real scene in virtual reality.The simulation proved that the combination of remote sensing images and digital elevation model can create a real road three-dimensional scene to meet the needs of driving simulation.This paper partly solves the requirements of high immersion and high operability in the road scene modeling of driving simulation.The research results can also be applied to intelligent driving,driver training,traffic simulation and highway evaluation,which has certain economic and social significance and a wide range of application prospects.