Research On CAD Model Based High Accuracy Pose Estimation Algorithm Of 3C Component

3C is the abbreviated name of computer,communication and consumer electronics products.Automated assembly of 3C production lines can effectively improve the labor environment of workers,reduce the labor intensity of workers,improve product quality,and reduce human error.The pose estimation of the assembly target is the basis of 3C assembly automation.The pose estimation based on point cloud registration has the advantage of non-contact and simple configuration,and is widely used in computer vision fields such as 3D reconstruction.However,there are not many researches on the pose estimation of 3C components with few features and high symmetry.Therefore,this dissertation uses 3C components as the pose estimation target,and uses the binocular structure light camera to complete the pose estimation task.In order to meet the industrial high-precision assembly requirement for 3C component,which is texture-less and highly symmetrical,a non-contact model-based pose estimation scheme are proposed.The scheme divides the pose estimation based on the model into two parts: global pose estimation and local pose estimation.The global pose estimation section quickly finds the initial pose of the 3C component relative to the camera,then local pose estimation section is used to improve the accuracy of the initial pose based on the global registration result.Considering the singularity problem caused by the high symmetry of the 3C component,in this dissertation,after the global pose estimation algorithm obtains multiple sets of initial poses,the template matching algorithm based on LINEMOD feature is used to eliminate the singular poses.In order to avoid the local pose estimation algorithm falling into local optimum due to insufficient accuracy of the initial pose,the BNB algorithm is combined with the ICP point cloud registration algorithm to perform spatial subdivision in the small space near the initial pose.After the split space is pruned,the ICP point cloud registration algorithm is used to finer pose in the remaining subdivision.Since the accuracy of the rotational space parameters of the ICP point cloud registration algorithm meets the requirements of the subject.We only performs spatial subdivision operations in the translation space,which reduces the spatial search dimension and improves the search efficiency.3C assembly automation needs to be completed by industrial robots.In this thesis,the calibration error of the pose between binocular structure camera coordinate and the robot base coordinate is taken into account,and the closed-form solution and the iterative solution algorithm are compared to select the best solution.Through simulation and actual experimental tests,the reliability and effectiveness of the whole 3C component pose estimation scheme are verified.The template matching algorithm based on LINEMOD feature can eliminate 100% singular pose,and the pose estimation accuracy based on BNB and ICP point cloud registration algorithm is higher than 0.3 mm.