| Simultaneous localization and mapping(SLAM)is a basic task in mobile robotics.The goal of SLAM is to estimate the pose of the robot through one or more sensors,simultaneously reconstructing the surrounding environment(map)in real-time.However,there are still many challenges in practical applications.For example,in the reconstruction of panoramic3 D maps,complex lighting,and low-texture wall scenes are all major challenges.In addition,GPU is usually required in the process of dense map reconstruction,which is difficult for mobile robots.However,the sparse map constructed by some current SLAM algorithms cannot meet the requirements of robot navigation and obstacle avoidance.Aiming at the above problems,this paper proposes a mobile robot localization and panoramic 3D map reconstruction algorithm based on CPU based real-time on binocular vision combining point and line features and IMU.The main contents of this paper are as follows:1.The research status of localization and map reconstruction of mobile robots at home and abroad is introduced,and the hotspots and difficulties in the field of SLAM are analyzed.The overall scheme design,hardware design and algorithm flow design of the mobile robot panoramic 3D map reconstruction and positioning system are introduced in detail.The sensors model used in this system are discussed,including the measurement model of the camera and the IMU sensor,the error model,and the calibration principle of each model parameter.Experimental comparison and brief analysis of sensor parameter calibration are carried out.2.For panoramic scene reconstruction and positioning that may encounter in the process of metope of low system for mobile robot localization and reconstruction in the scene to extract texture features of fewer problems,designed based on improved point line visual characteristics and innovative marketing combined with the front end of the odometer.It mainly includes the extraction and matching of improved point-line features and the construction of residual model,the calculation of IMU residual model,the simulation and visualization of Hessian matrix and the construction of prior residual model.Finally,stereo matching is performed on the image information of two consecutive keyframes to obtain the corresponding disparity image.Then,based on the relative pose relationship between keyframes,the disparity image is converted to a 3D point cloud in world coordinates,and it is meshed to generate a global 3D dense map for scene visualization.3.To solve the problem of slow initialization speed of existing SLAM systems,an improved joint initialization algorithm of vision and IMU was proposed,which used depth information to speed up the initialization when depth information was known.4.In order to ensure the real-time performance of the system,the solution algorithm of back-end nonlinear optimization was improved,and the LM algorithm of the third party optimization library was replaced by Dogleg's algorithm attribute to improve the solution speed.At the same time,in the panoramic 3D reconstruction module,the improved SLIC algorithm is used to reduce the dimension of the reconstructed data,so as to speed up the back-end reconstruction and reduce the operation force of the platform.Finally,the data set and online performance test of the mobile robot's panoramic 3D map reconstruction and positioning system are carried out.The root mean square error of positioning accuracy can reach 0.033 m in the ordinary scene of Eu Roc dataset,and the root mean square error of positioning accuracy can reach 0.073 m in the scene of TUM-VI corridor low texture dataset,and the average tracking frame rate is 25 Hz.The system's actual scene test positioning accuracy can reach 0.0214 m,and the scene reconstruction error is0.12%?0.17%.Through experimental verification,the algorithm in this paper can perform stable positioning and scene reconstruction in both ordinary scenes and low-texture scenes. |
PDF Full Text Request