Design And Research Of Indoor 3D Reconstruction And SLAM System Based On Binocular Vision

In mobile robot technology,accurate sensor information is essential for automatic positioning and meaningful navigation.In order for the robot to perform tasks autonomously,it needs to obtain accurate information on the surrounding environment and position,namely,localization and mapping(Simultaneous localization and mapping,SLAM).SLAM is a very important issue in the research of robot technology.It uses attitude information and visual odometry to estimate the trajectory of the robot's movement and construct a map of its surrounding environment to achieve autonomous control and navigation in an unknown environment.In SLAM,the interaction between map construction and positioning is crucial:if the positioning is wrong,the new sensor information in the existing map will be inconsistent,causing drift.In the process of positioning based on the visual sensor,the posture change and displacement during the camera movement can be solved according to the multi-view geometric relationship of the current frame and the previous frame,and the multi-view between the camera posture information and the image information on different time axes Geometric relationship to achieve three-dimensional reconstruction of indoor scenes.Solving SLAM problems usually depends on the type of sensor information available and the limitations of application scenarios.Among them,laser SLAM research is relatively mature and robust,and is the current mainstream navigation method;visual SLAM has low cost,rich information,and wide application range.The advantages such as have received extensive attention,but the disadvantage is that they are very dependent on the characteristic information of the surrounding environment,and are easily affected by ambient light.Based on this,this paper studies and designs an indoor 3D reconstruction and path planning system based on binocular vision.The main research work is as follows:1.The accuracy of the three-dimensional reconstruction of the camera depends on the accuracy of the camera model and the accuracy of the image depth information,and is affected by the ambient light and the indoor environment.This paper systematically analyzes the camera imaging model,and performs parameter calibration and image correction for the binocular camera.In the process of feature point detection and matching,an improved method based on FLANN matching algorithm and RANSAC to eliminate mismatched points is proposed,which can effectively improve the accuracy of matching.2.In view of the problem of large data volume and slow processing speed in the past3 D reconstruction process,this paper builds a 3D reconstruction system equipped with Nvidia TX2 processor based on the ORB-SALM2 algorithm.Compared with the use of traditional CPUs,Nvidia TX2 can greatly improve the image processing speed,run more efficiently,and meet the real-time requirements.3.In the process of solving the camera movement trajectory,it takes a long time,and it is easy to lose frames and accumulate errors.Based on the SGBM method,this paper designs a semi-global stereo matching method.It solves the occlusion detection,sub-pixel refinement,and multi-level baseline matching during image matching,which takes less time than global matching.4.Finally,this paper builds the SLAM system experimental platform based on the Mecanum wheel,equipped with binocular camera and IMU sensor,realizes the binocular 3D reconstruction process,and experiments the ORB-SLAM2 algorithm to achieve indoor positioning and path planning,The feasibility and accuracy of the system are verified.