Reconstruction Of 3D Objects Based On RGBD-SLAM

With the continuous development of robotics and computer vision theory,Simultaneous Localization And Mapping(SLAM)technology has been widely used in many fields such as robot positioning and navigation,VR/AR,human-computer interaction,automatic driving and scene reconstruction.Among them,the 3-D reconstruction method based on SLAM can achieve the reconstruction requirement through the real-time estimated pose of the SLAM system.However,this method has high requirements on the hardware level and generally needs the help of the GPU.At ordinary CPU level,dense point cloud based on SLAM is difficult to achieve real-time effect,and point cloud quality is difficult to reach a very high level due to the lack of accuracy of pose estimation.Compared with the traditional GPU-based approach,this paper designs and implements a comprehensive SLAM-based offline optimization and point cloud reconstruction and extraction system at the CPU level.This paper mainly focuses on the following three parts:1.Design and implementation of a KeyFrame-Based RGBD-SLAM system.This section is a regular visual SLAM system that records the image and pose of the key frame,and loop information for offline optimization.2.Design and implementation of an offline optimization and 3D point cloud reconstruction system.This part takes the information recorded by the SLAM system as input,optimizes the pose of the key frame offline,and reconstructs the map point cloud.3.Design and implementation of an interactive point cloud separation and extraction system.This part is presented in the form of QT software.After the map cloud point is completed,the target point cloud can be accurately extracted according to the needs of users.To sum up,the integrated system designed and implemented in this paper is actually a combination of KeyFrame-Based RGBD-SLAM system,offline optimization and 3D point cloud reconstruction system and interactive point cloud separation and extraction system.Through the accuracy experiment of the point cloud reconstruction,the performance and function of each key part in this paper are verified.This article also compares with other advanced SLAM systems in the open dataset,and proves that this method has good accuracy and reliability,and obtains satisfactory results.