Monocular Vision Pose Measurement Of 3D Object Based On CAD Model Library

At present,vision-based pose measurement has been widely used in industrial automation.With the rapid development of modern industry,the requirement for sensory ability of robots is constantly increasing,and vision-based pose measurement plays an important role in improving the intelligence of robots.Monocular vision system uses only one camera,which has simple structure and relatively low cost,and is widely used.At present,pose measurement methods based on monocular vision are mainly divided into two categories:based on object features and template matching.The method based on object features usually uses geometric features such as points,lines and circles to match,sometimes it needs to set feature points artificially.The template matching method establishes the template library offline according to the image of the object's CAD model in various postures.Then the measured image is processed and matched with the image in the template library to determine the current pose of the object.Because the method based on object features has limitations in application scope,this master's thesis focuses on the idea of template matching,mainly launches the research of monocular vision 3D object pose measurement based on CAD model library.This research topic has good research value and application prospects.The main work and innovation of this master s thesis are as follows:Firstly,to solve the problem of pose measurement of 3D objects under monocular vision,a monocular camera pose measurement method based on improved ICP algorithm and CAD model library is proposed.First of all,OpenGL is used to draw the geometric model of the object,a series of virtual observation points are set up to sample the projection of the object under the virtual camera,and the set of edge points of projection images are extracted to build the template library.Then,the improved Hausdorff distance is used to match the edge point set of the image to be processed with the point set of the image in the template library,and the rough pose is obtained.On this basis,the improved ICP algorithm is used to iteratively optimize the edge point set of the measured image and the 3D edge point set of the object to obtain accurate pose parameters.Secondly,to solve the problem of limited sampling spacing in CAD model base,a monocular vision pose measurement method based on multiple template images is proposed.The edge points set extracted from the measured image is matched with the edge points set of the template image,and eight template images with the smallest Hausdorff distance are obtained to ensure that the pose of the object in the measured image is within the range represented by the eight template images.On this basis,the improved ICP algorithm is used to iterate the edge points set of the measured image and the edge points set of the eight template images to obtain eight groups of optimized pose parameters,which are given different weights to calculate the final pose.Thirdly,aiming at the velocity of pose measurement for 3D objects,a new monocular vision pose measurement method based on image pyramid and improved genetic algorithm is proposed.Based on the improved Hausdorff distance and image pyramid,the rough pose is obtained by matching the measured image with the template image,which reduces the matching time.On this basis,the improved genetic algorithm is used to iteratively optimize the rough pose parameters to obtain the optimized pose parameters of the 3D object.Fourthly,based on the application of vision system in industrial automation,a large-scale workpiece automatic loading and unloading system based on vision system and industrial robot is proposed and implemented by combining pose measurement with robot hand-eye calibration.The system includes two main processes:off-line calibration and teaching,on-line pose detection and grabbing,the experimental results show that the proposed automatic loading and unloading system for large-scale workpieces can achieve high-precision pose measurement and grabbing,and meet the needs of industrial automation applications.