The Research Of Dynamic Image Stitching And Reconstruction Technology For Environmental Perception Of Vehicle

The technology of dynamic image stitchingand reconstruction forenvironmental perception of vehiclethat isdetect pedestrians and obstacles around the vehicle in the urban road and generate a panoramic map of vehicle running environment to help drivers better understand the vehicle running environment.Sequence images with overlapping parts are mainly obtained by in-vehicle camera,and sequence imagescan be reconstructed 2D or 3D environment around vehicle by image stitching and reconstruction technology.The panoramic map of vehicle running environment is generated to help drivers better understand the vehicle running environment and realize the safety of driving.The research of image stitchingand environmental reconstruction technology based on visual sensor has become a research hotspot in the field of intelligent vehicle and has important theoretical significance and application value.This thesis studies on the stitching algorithm of sequence images and the reconstruction technology of vehicle environment.A stitching algorithm of sequence images based on determination of plane region and a 3D environment reconstruction algorithm based on segmentation of plane region are proposed in this paper.And a sequence images stitching and reconstruction system for vehicle environment perception is designed.The perception of the surrounding environment of vehicle and t the position measurement of vehicle in the environment have been realized.The simulation experiments and the real car tests validate the advantages and effectiveness of the proposed algorithm.The research work of this paper is as follows.This thesis reviews the research background and the development of dynamic image stitchingand reconstruction technology forenvironmental perception of vehicle.And this thesis analyses the problems of image stitching and reconstruction in vehicle environment perception and presents the research objectives of this project.In view of the problems of the existing dynamic image stitching algorithm,such as large amount of computation and the effect which is difficult to meet the actual requirements,according to the actual requirement of vehicle environment perception under the urban road environment and the particularity of the feature points in two-dimensional pavement image,the distance invariance of the motion of vehicle coordinate system between the plane feature points is studied under the inverse projection transformation of the plane region.And the method for determining the plane region of vehicle environment is proposed.On this basis,a fast algorithm of vehicle motion matrix estimation based on triangulation matching of image feature points in plane area is proposed,which realizes fast matching of vehicle motion images and the two-dimensional environment reconstruction of vehicle.Simulation experiments verify the effectiveness and real-time performance of the proposed algorithm,and the actual requirements of vehicle environment perception can be satisfied.For the relative stable plane region characteristics of in-vehicle camera sequence images,a fast algorithm for 3D reconstruction of vehicle environment based on plane region determination and segmentation is proposed in this paper.The algorithm utilizes the feature points on the image plane region to estimate vehicle motion,direct inverse projective transform of the planar region image and generate two-dimensional image.And the non-planar image region utilize SLAM(simultaneous localization and mapping) technology to generate 3D point cloud.The simulation results show that the algorithm effectively improves the accuracy of vehicle environment perception and the effectiveness and applicability of the algorithm.This thesis designs a software platform for dynamic image stitchingand reconstruction forenvironmental perception of vehicle,and the relevant algorithm proposed in this paper is applied to real vehicle test to verify the effectiveness of the algorithm in engineering application of vehicle environment perception.This thesis can establish a good technical foundation for the design of Advanced Driver Assistant System(ADAS) for intelligent vehicle.The research work is summarized,and the future prospect of the research is given,which will provide new ideas for further research.