Research On Application Of Cloud Registration Algorithm In Stacked Parts Picking

In the actual process of point cloud collection,due to the changes in the subject’s viewpoint,the obstruction of interfering objects,and the perspective constraints of sensors,the same object often needs to be collected multiple times,resulting in inconsistent geometric centers of point cloud data.The research conducted to solve such problems is called the research on point cloud registration algorithms,which is of great significance for machine vision guided automation of high-end equipment.This dissertation focuses on the problem of picking stacked parts,taking the registration part in the point cloud recognition process as the starting point,and focuses on solving the problem of decreased registration accuracy caused by prior assumptions.A particle swarm optimization based registration algorithm is proposed,which dynamically optimizes the registration threshold parameters through particle swarm optimization,achieving adaptive threshold parameters during the registration process and improving the registration accuracy of part point clouds.And the effectiveness of the algorithm proposed in this paper in improving the recognition effect of stacked parts was experimentally demonstrated.The main research content of this paper is as follows:(1)Firstly,KD tree algorithm is used to build the topological space of point cloud,improve the query speed of point cloud,and realize the K neighborhood query of points.By using statistical filters to eliminate outliers and eliminate clutter interference,voxel downsampling method is used to simplify point cloud data.Principal component analysis is used to calculate the normal vector of the point cloud,and then the point cloud is preliminarily screened using its curvature change characteristics.This method utilizes curvature features to perform secondary filtering on point clouds while excluding noise interference,effectively reducing the number of point clouds while preserving their geometric features.(2)A particle swarm optimization based registration algorithm is proposed to address the issue of decreased registration accuracy caused by prior assumptions in classical registration algorithms.This algorithm builds an adaptive registration threshold structure based on particle swarm optimization,and guides the changes in thresholds such as nearest neighbor ratio,geometric consistency,and minimum distance in feature matching and mismatching removal through objective functions.It iteratively obtains the global optimal geometric feature measurement threshold,thereby improving the accuracy of feature matching.The Stanford University dataset registration comparison experiment shows that compared to the RANSAC algorithm,the algorithm has an average matching rate improvement of 27.9%,an average MAE improvement of 32%,and an average MSE improvement of 1.64% on different geometric types of point cloud data,which can achieve better matching results.(3)In response to the problem of low recognition rate caused by occlusion and noise during classical point cloud recognition.This article designs a neighborhood based approach based on the combination of root mean square error and point cloud disorder features＿Root mean square error and cross entropy The objective function of the root mean square error of the neighborhood.According to the experimental results of the objective function comparison,the cross entropy with the best registration effect is selected The neighborhood root mean square error as the objective function effectively improves the robustness and accuracy of the registration effect of the algorithm.Through comparative experiments on part dataset recognition,it is shown that compared to the RANSAC algorithm,the average number of iterations of this algorithm is reduced by37.8%,and the RMSE is increased by 9.98% on average.It is superior to the RANSAC algorithm in recognition time and accuracy,effectively improving the efficiency of part pose recognition.