The Design And Implementation Of Reconstruction System For Unknown 3D Scene

With the widespread using of RGBD cameras and the significant progress in scene reconstruction recent years,the high quality of indoor scenes scanning and high-density 3D reconstruction of mobile robot is paid more and more attention on the robotics and graphics.In 3D scene reconstruction,human beings can quickly build the shape of the object in the scene according to the ability of vision.However,due to the complexity of the visual problem itself,the variaty of scene information and the complexity of computing,current robot visual system is difficult to achieve the cognitive ability of human.The relationship of scene objects in the space is complex,the occlusion will case the loss of data,and the complex relationship between will lead to the difficulty of segmentation.Moreover,the appearance of objects is complex and the diverse texture,the change of light condition will lead to the difficulties of recognition,and complex geometry will also affect the reconstruction effect.The reconstruction of unknown scenes is usually driven by the spatial information of the scene.Therefore,the key to the reconstruction of an unknown 3D scene is the realization of the robot's automatic navigation to obtain the range information of scene.The high-quality scan used for scene reconstruction depends on the local geometric information of the visible surface,which makes it necessary for the robot to move accurately to ensure complete and stable scanning,as well as to be able to acquire consecutive frames and reduce scanning and reconstruction errors.The realization of the rapid scanning of the environment and the balance of high-quality scene reconstruction are the main challenges for the autonomous reconstruction of the scene.In order to enable robots to have a better understanding of unknown 3D scenes,this paper proposes an unknown 3D scene reconstruction method based on path planning.Online 3D reconstruction is based on voxel fusion method.The progressively reconstructed 3D scene has constraints and guidance for the walking of the robot and the next scanning,and the path planning is performed through the 2D field.Both the smoothness of local movement and the efficiency of exploration are taken into account.The ultimate goal is to use high-quality path planning for high-quality 3D reconstruction.