Research On Parts Identification And Positioning For Robot Crawling

With the widespread use of industrial robots,the degree of industrial automation has been greatly enhanced.Industrial robots play an important role in the auto assembly process of industrial production.Compared with the traditional manual assembly,robotic assembly based on industrial robots has obvious advantages in terms of efficiency,accuracy and flexibility.At present,the application of industrial robot in automatic assembly still has many deficiencies.In most domestic applications,robots still need some artificial assistance or could not complete fully automated assembly.The main technical bottleneck is the recognition and localization of the target when the robot grabs the target.In this paper,aiming at the problem of picking up parts in robotic automatic assembly,a method of component recognition and location based on binocular structured light stereo vision is proposed.The main research contents are as follows:Firstly,the overall program of part identification and location is researched.According to the structure information of the parts to be picked up,the environment,the state of picking up the parts,the requirements of picking up and so on.The scheme of the part recognition and positioning based on the binocular structured light stereo vision is proposed.We set up a visual measurement hardware system,and developed a corresponding measurement,identification,positioning software modules.Proposed a two-dimensional image recognition and 3D point cloud positioning overall program.Secondly,for the multi-target recognition problem of stochastic stacking,several common target recognition methods in images are analyzed.The HOG-based SVM target recognition classifier is studied.According to the deficiency of existing methods,a HOG feature based on feature region and multi-scale target recognition scheme based on image pyramid are proposed.The positive samples were generated by sampling the parts under different poses,sub-images are randomly taken from the scattered pile of parts of the image as a negative sample to train support vector machines.The experimental results show that the SVM classifier trained by thisscheme can effectively detect the target parts in the picture with high detection efficiency and reliable test results.Then,based on the recognition result,the measurement scene is partially reconstructed in 3D.Finally,we studied the part positioning problem based on 3D point cloud.The result of reconstruction is the point cloud of a single part in the scattered parts.Aiming at the pose estimation of parts,a method of locating target parts based on geometric primitives is proposed according to the structure of parts.The geometric primitive fitting algorithm based on RANSAC algorithm is studied.The local coordinate system is established according to the result of primitive fitting and the positional relationship of primitives.By matching the local coordinate system and calculating the conversion relation between the local coordinate systems,the pose of the target part relative to the CAD model is obtained.