Research On Indoor 3D Reconstruction Algorithm Based On Panoramic Image

Three-dimensional reconstruction of indoor scenes is an important research topic in the field of computer vision.Efficient,low-cost,accurate 3D reconstruction algorithm of indoor scenes has a wide range of applications in virtual reality,augmented reality,indoor positioning,robot navigation,and so on.The purpose of three-dimensional reconstruction of indoor scenes is to make computer understand the three-dimensional geometric relationship of the scene and to obtain the three-dimensional spatial structure of the scene by acquiring the depth information of the scene.The existing depth estimation methods and scene reconstruction algorithms either rely on expensive depth acquisition equipment or have complex algorithms and spend a long time on calculations,which increase the application threshold of 3D scene reconstruction technology.In view of the limitations of the existing 3D scene reconstruction technology,this thesis studies the problem of indoor 3D reconstruction based on panoramic images,and the main work is as follows:1.This thesis proposes a three-dimensional reconstruction algorithm of indoor scenes based on a single panoramic image to efficiently recover the three-dimensional layout model of indoor scenes from the panoramic images combining the structural characteristics of indoor scene and the projection relationship of panoramic images.Firstly,the indoor scene is used to satisfy the manhattan world assumption.The algorithm extracts the line information in the image by combining the constraints obtaining the outline of the indoor scene.Secondly,the algorithm uses symmetry to map the ground contour line segment from the ceiling contour line segment solving the problem that the object blocks the ground contour line.Finally,the layout assumption of the indoor scene is obtained by dividing the image by the contour line segment in the panoramic image;using the projection relationship of the panoramic image,the pixels of the wall,ceiling and ground areas are converted into three-dimensional point clouds to reconstruct the layout model of the indoor scene.The experimental results show that the reconstruction algorithm proposed in this thesis can efficiently reconstruct the layout model of the indoor scene through a single panoramic image,and the input data is easy to obtained.2.This thesis combing the mobile phone sensor data to achieve indoor positioning,obtains the attitude data when the panoramic camera moves,and calculates the translation matrix and rotation matrix between different shooting viewpoints through pedestrian dead reckoning,to achieve automatic splicing of different room layout models in the same scene to improve the applicability of the reconstruction algorithm.