| Intelligent manufacturing is inseparable from the development of industrial robot technology,and workpiece sorting is an important application of industrial robots.Adding 3D technology to industrial robots can not only increase the flexibility of the system,but also liberate part of the labor force.It is of great significance for industrial intelligent manufacturing to improve the system's ability to automatically recognize and estimate the pose information of the workpiece,which is therefore the main research purpose of this thesis.In order to solve the problem of insufficient correction when using monotone method to correct the phase noise of shadow area in binocular vision system,an improved phase correction method is proposed.The absolute phase information of the workpiece surface is solved by using four-step phase shift and multi-frequency heterodyne,and the phase noise is corrected by monotonic non-decreasing;The phase noise characteristics of multiple groups of different experimental subjects with insufficient correction are analyzed,and the discrete phase noise is corrected by calculating the frequency of non-zero phase in the fixed interval according to the phase frequency matrix.The continuous phase noise is corrected by using the difference between the actual phase and the predicted phase of the adjacent discontinuity points;The slope between the discontinuities and the non-zero phases on both sides are calculated and judged,and linear method is used to compensate part of the phase which is set to zero because of random noise;The epipolar correction is completed according to the calibration data of the binocular camera,and the fast phase matching is completed on the constraint conditions such as sequence consistency.The results show that the proposed method can reconstruct wooden cylinders,wooden boxes,metal cylinders and plastic bottles.The standard deviations of the corresponding triangular plate model is 0.0978 mm,0.0250 mm,0.1070 mm and 0.0944 mm respectively,the average time is 7.5s,and the average point cloud number is 1407857,which shows that the proposed method can be applied to different workpiece objects.Aiming at the high time-consuming problem resulted by the large scale of workpiece point cloud data,a point cloud registration method based on down-sampling optimization of key points is proposed.The background points and outliers of the workpiece point cloud are removed by using the through filter and radius filter,respectively.Multiple workpieces are divided into a single workpiece point cloud by using the clustering algorithm based on Euclidean distance;The center of gravity of several voxels in a single workpiece point cloud is calculated,and kd-tree is used to quickly traverse the neighboring points of the center of gravity to replace the voxels,thus realizing downsampling;An adaptive point cloud average distance calculation method and a boundary point judgment method based on spherical neighborhood are proposed to optimize the key points of ISS;The optimized key points are described by FPFH features,the approximate transformation matrix is solved by SAC-IA,and the precise registration is realized by ICP algorithm.The results show that compared with the four registration algorithms of ICP,ISS+FPFH feature registration,ISS+FPFH+SAC-IA,ISS+FPFH+SAC-IA+ICP,the registration accuracy is improved by 96.9%,98.1%,93.3% and 3.5%,and the registration speed is improved by 77.2%,77.7%,76.9% and 85.4%,respectively.In order to verify the effect of algorithm pose estimation,an experimental platform is built.The multi-frequency heterodyne is utilized to reconstruct the stacked plastic bottle workpiece and the stacked wooden cylinder workpiece respectively,and the point cloud of the workpiece is preprocessed.A point cloud registration method based on key point optimization after down-sampling is used for registration,and the pose estimation of each workpiece is performed respectively.The results show that the algorithm is feasible in random bin picking technology. |
PDF Full Text Request