Research On 3D Reconstruction Of Indoor Scene Based On Monocular Camera

With the rapid development of the robot industry in recent decades,more and more service robots and industrial robots have been widely used in human society.As people have higher requirements for the intelligence of robots,3D reconstruction technology,which is the key technology for robot environment perception,has become a hot research direction and been receiving more and more attention.SLAM(Simultaneous Localization and Mapping)is an effective tool for 3D reconstruction.In a completely unknown environment,the robot obtains environmental data through its own sensors to build a map consistent with the real environment,and at the same time determines its pose in the environment by itself.And the monocular vision SLAM has attracted the attention of researchers due to its low cost and wide application scenarios.Binocular stereo vision is a 3-D reconstruction algorithm based on two-view geometry.The algorithm steps include camera calibration,distortion correction,stereo correction,stereo matching,and obtaining depth maps through disparity maps.In this paper,3-D reconstruction is carried out on the basis of monocular vision SLAM algorithm and binocular stereo vision algorithm.Aiming at the lack of scale in monocular vision SLAM and the accuracy of binocular stereo camera triangulation.First,the size information and control information of the robot are combined to form the equivalent baseline of the binocular camera to achieve the purpose of the monocular camera equivalent to the binocular camera.While solving the problem of the lack of 3-D reconstruction scale of the monocular camera,a variable baseline strategy is adopted to achieve the purpose of improving the accuracy of triangulation.At the same time,the traditional pose estimation algorithm is analyzed.Aiming at the problem that it cannot completely eliminate noise points,the method of matrix low-rank sparse decomposition is used to reduce the influence of noise points and improve the accuracy of camera relative pose estimation.In addition,an active closed-loop optimization algorithm is proposed combining the ideas of loop detection and pose map optimization.The prior information provided by the robot control system and size information is used to make the camera achieve the effect of active loopback,and the improved relative pose estimation algorithm is used to adjust the overall situation.The pose energy function to achieve a better global pose estimation effect.This paper uses the data collected by sensors in the physical simulation environment for experiments and analysis.The experimental results show that the algorithm proposed in this paper can not only improve the accuracy of depth estimation,but also the accuracy of relative pose estimation.At the same time,the global pose optimization effect is improved,which proves the correctness of the improved algorithm in this paper.