| The detection and pose estimation of objects are an important basis for workpiece sorting and assembly,and is also a crucial part in robotic grasping.Therefore,it has become a research hotspot at home and abroad.Based on the detection and pose estimation task of the lock pin workpieces,this thesis studies and improves the detection and pose estimation algorithm based on point pair feature(PPF)approach.The experimental analysis on UWA dataset and workpiece data respectively verifies the effectiveness of the improved algorithm.The main content is as follows:(i)The technology of point cloud acquiring and preprocessing are studied.According to the vision system and camera calibration principle,the depth image and the scene point cloud are collected.Aiming at the noise in the depth image and point cloud,it is filtered through bilateral filter and statistical filter to improve the quality of point cloud.(ii)In order to improve the recognition rate and running speed of the detection and pose estimation algorithm based on PPF,our improved methods mainly include: In preprocessing,this thesis improves the point cloud downsampling strategy through the normal cluster voxel grid method,which not only remains discriminative characteristics of the underlying surface,but also reduces redundant data.In matching and voting,this thesis optimizes the Hough voting strategy by calculating the voting weight of PPF,which reduces the influence of similar points mismatching on pose estimation.In postprocessing,a re-scoring strategy is proposed to better quantify the fit of the model to the scene.Finally,experiments and analysis are carried out on the UWA dataset and workpiece data respectively.The results show that the improved algorithm effectively improves the recognition rate and running speed.(iii)The construction of PPF is studied.Aiming at the problem that the ambiguity of the normal direction easily causes the mismatch of PPF,the minimum spanning tree algorithm is used to unify the normal direction of the model and the scene.In view of the high complexity of point pair feature calculation in scene point cloud,the method of limiting the point pair distance is used to reduce the number of PPF in the scene point cloud. |