Object 3D Reconstruction And Pose Estimation

With the declining birth rate and the increasing population aging problem,my country's demographic dividend is rapidly disappearing.How to realize intelligent manufacturing as soon as possible and complete the industrial upgrading of the manufacturing industry has become more and more important.Robots are important to realize intelligent manufacturing.Three-dimensional reconstruction and object 6D pose estimation are the key to the vision technology in operating robots.Among them,the speed and accuracy of the object 6D pose estimation algorithm directly determine the success or failure of the operation task of the operating robot,and the three-dimensional reconstruction provides a model for the object 6D pose estimation algorithm.The data makes it possible to estimate the 6D pose of the object.Based on this,this paper has carried out the research of object pose estimation and 3D reconstruction in the robot operation scene.The main work includes the following three aspects:1)Aiming at the problem of the lack of a 3D model for the 6D pose estimation of the object in the operating robot scene,this paper builds a fast 3D reconstruction system for the object based on the turntable.The system is mainly composed of a depth camera,a color camera and a rotating turntable with a controllable angle.In order to reconstruct a high-precision three-dimensional model,this paper calibrates the color camera,the depth camera,and the turntable.The experiment shows the calibration accuracy meet the needs of 3D reconstruction.2)The design implements a fast 3D reconstruction method for objects based on the turntable,in which the rotating platform can be rotated around the central axis of the rotating turntable at a preset angle,and the target object can be collected by adjusting the posture of the object on the turntable multiple times and the calibrated depth camera and color camera.The multi-view color point cloud data is registered and fused,and the surface of the object is reconstructed using the Poisson surface reconstruction algorithm to obtain a complete three-dimensional model of the object.Experiments show that the reconstruction accuracy of this method can reach 0.42 mm.3)The point-to-feature algorithm uses the 3D model reconstructed by the turntable-based object rapid reconstruction system and the scene point cloud data collected by the depth camera to perform 6D pose estimation.First,perform global modeling of the three-dimensional model of the object and store it as a hash table to establish a point-to-feature library,and then calculate the point-to-feature index hash table for the scene point cloud to find the corresponding point to calculate the pose,and select the best pose by voting.The optimal pose,and finally the pose is further corrected by the iterative nearest point algorithm.Experiments show that the pose estimation accuracy of this method can reach 1.51 mm,which meets the grasping requirements of the robotic arm.