Font Size: a A A

Virtual Scene Reconstruction For Intelligent Driving

Posted on:2020-03-07Degree:MasterType:Thesis
Country:ChinaCandidate:H Y WangFull Text:PDF
GTID:2392330596475943Subject:Control Science and Engineering
Abstract/Summary:PDF Full Text Request
As a key technology of the new generation of artificial intelligence,intelligent driving involves multi-disciplinary knowledge such as environment perception,mobile internet and big data processing.Among them,the use of virtual reality technology to achieve intelligent system performance simulation,not only can test and verify the intelligent system hardware and software functions,reduce development costs,but also greatly improve the user's sensory experience,and further deep system development,because of its interactivity and conception.This paper is aimed at intelligent driving system,using computer graphics rendering technology to carry out research on virtual reconstruction of 3D real scene,and proposes two sets of image-based 3D scene rendering methods,and its implementation scheme on semi-physical driving simulation platform,aiming to effectively improve the smart driving simulation system displays realism.The main research works are as follows:Aiming at the display distortion problem of traditional 3D virtual reconstruction technology,this paper proposes a 3D real scene virtual reconstruction method which can reduce the display distortion degree by taking the driving record panorama or video as input.The distortion problem of traditional three-dimensional virtual reconstruction display is caused by the superposition of the geometrical distribution of the spherical spheres of the panoramic picture/video carrier sphere.For the problem of uneven geometric distribution of the spherical body in the scene,the "weft loop method" QTM model is used to generate the scene-carrying sphere.A low distortion material mapping method is proposed to generate a sphere with highest uniformity which is used for carrying scene.The method performs an accurate texture mapping on the sphere with the input panoramic picture/video.Finally,a three-dimensional real-time reconstruction method is obtained that effectively reduces the degree of display distortion,and 360° omnidirectional viewing of panoramic driving records is realized.In binocular computer vision,in order to simulate different viewpoint scenes and render the visual effect of vehicle lateral displacement.Based on the halfway domain view morphing algorithm and the driver's visual focus distribution information,this paper proposes a view morphing algorithm using ORB feature matching information.The view morphing algorithm is specially optimized for the intelligent driving simulation system,which not only has better image output quality than various classical view morphing algorithms,but also has faster execution speed than the halfway domain view morphing algorithm.Finally,this paper proposes an implementation scheme of image-based 3D scene reconstruction rendering method for intelligent driving simulation system,and completes the interactive system software/hardware configuration to realize the proposed algorithm.The solution uses the Unity 3D engine platform and accurately maps the control device commands on the intelligent driving simulation platform to the actual actions of people and vehicles during the driving simulation process through the built-in controller mapping function.The complete coverage of the two scene reconstruction methods proposed in the paper from theory to practical application is realized,which fully reflects the theoretical and applied value of the research work.
Keywords/Search Tags:Intelligent driving, Virtual reality, Computer graphics rendering, View morphing
PDF Full Text Request
Related items