Research On 3D Reconstruction And Motion Estimation Methods Based On Structured Light Vision

In recent years,with the rapid development of computer vision technology,more and more experts and scholars attach importance to 3D reconstruction.Along with the increasing complexity of 3D reconstruction scene,the difficulty of reconstruction is also increasing.The occlusion of dynamic objects in static scenes,the occlusion between point clouds and environmental noise will affect the reconstruction accuracy.In recent years,the related methods are divided into two categories: one is for the 3D reconstruction of fixed static scene,the other is for the dynamic object to be placed on the turntable,using the static camera to capture the image and obtain the point cloud data under different views.In this paper,we will discuss the reconstruction of the dynamic rotating object under views of a moving RGB-D camera,and further estimate the motion information of the moving object.This paper first analyzes the pinhole imaging model and the mathematical model of the structured light vision system.Then the image preprocessing module is introduced which is divided into two modules: object tracking and object segmentation.By using object tracking,the region of the object is reduced to improve the accuracy of object segmentation.The object tracking module adopts KCF tracking algorithm,while the object segmentation module adopts Grab Cut algorithm.The ROI matrix obtained from KCF tracking can be directly imported into Grab Cut algorithm,then the whole preprocessing module can be integrated.After image preprocessing,RGB-D image data can be synthesized into 3D point cloud data.Considering different densities of texture and geometric features of different objects,this paper uses weighted 2D SIFT feature points and 3D FPFH structure features to perform initial registration of adjacent frames,and further uses the iterative closest point(ICP)algorithm to achieve accurate registration.Then the closed-loop optimization module is used to eliminate the accumulated error and obtain the complete point cloud model of the target object.After getting the optimized global pose,the corresponding motion information can be calculated.In order to further improve the accuracy and reconstruction quality of the model,this paper realizes the 3D dense reconstruction model based on TSDF model.Finally,mask information is used to realize the camera pose calculation removing moving object,which can be used to recover the motion information.By integrating the above modules,this paper completes the three-dimensional reconstruction and pose estimation based on the dynamic object scene.Finally,in order to verify the reliability of the reconstruction system,the algorithm used in this paper is compared with other three-dimensional reconstruction and pose estimation algorithms.Meanwhile,the algorithm proposed in this paper is verified by rotating platform and structured light camera.The given experimental results also verify the effectiveness and feasibility of the algorithm proposed in this paper.