Research On Depalletizing Analysis And Localization Method Of Mixed-loaded Pallet Based On Three-demensional Vision Technology

At present,more and more mixed loaded palletizing technology are used in logistics to form a mixed pallet,in order to allocate a proper amount of labor force to reduce supply chain costs and transportation costs.As an indispensable part of the logistics,Due to the characteristics of the mixing pallet structure,the depalletizing task of the mixed-loaded pallet demands more from the traditional depalletizing technology in the aspects of flexibility and accuracy.In this thesis,combining with 3D vision technology,a target depalletizing analysis and positioning method is proposed based on laser trangulation measurement,which solves the problems caused by the disorderly stacking of multiple different sizes of boxes in the mixedloaded pallet.At the same time,the interference caused by the strong reflective material contained in the actual mixed-loaded pallet on the three-dimensional reconstruction is eliminated.At first,the composition and principle of the laser triangulation measurement system are studied,and establishes a three-dimensional reconstruction calculation model.According to the characteristics of the mixed-loaded pallet,the basic parameters of the vision system and the hardware layout are determined.Then the hardware selection and parameter calibration of the vision system are completed.The relationship between the depth error of the three-dimensional reconstruction of the vision system,the depth of field and the working height are analyzed.Finally,the hardware design of the vision system experiment platform is completed.For the material with strong reflection such as tape existing in the actual mixedloaded pallet,the laser line image is preprocessed by using the optical filter and the filtering method,and the region of the laser line is filled in combination with the morphological processing,this method can effectively suppress or eliminate the interference generated by the strong reflection in the three-dimensional reconstruct vision system and provide high-quality laser line images for laser triangulation.Then 3D reconstruction is completed on this basis.For the problems in the reconstructed 3D point cloud data such as the point cloud of invalid target and large point cloud density,the pass-through filtering and voxel mesh down sampling method are used in preprocessing stage,and the valid target point cloud is maintained and the number is reduced.Using the region-based point cloud segmentation method for the threedimensional data on the surface of the mixed-loaded pallet,the geometrical characteristics of point cloud on the surface of mixed-loaded pallet,the angle between the local normal vectors of the point cloud and bending degree are combined as the similarity measure to complete the segmentation.Then,a method for judging the target detachability independent of template matching is proposed.The method makes full use of the point cloud spatial occlusion relationship after three-dimensional reconstruction,and uses the proximity of the boundary of each target point cloud cluster to judge the existence of occlusion.For relatively close targets,the occlusion relationship between the adjacent cluster is distinguished according to the depth values of adjacent points on the boundary.Finally,under the Ubuntu14.04 LTS system,the development of the ROS-based software package is completed with integration of the HALCON machine vision library.The experimental verification of various test samples is performed.The vision depalletizing method has a good effect on the detachability judgment of the mixed-loaded pallet and the calculation of the target pose,and has certain adaptability in the application of the mixed-loaded depalletizing.