Visual Positioning Technology For Naturally Stacked Work-pieces

Intelligent,automated robot technology is the future direction of development.Stereo vision plays a crucial role as the visual guiding part of the robot.Stereoscopic vision is a guidance that makes robots capable of performing tasks of all kinds,like grabbing the work-pieces,autonomous loading and unloading of parts,and so on.Since there is no human interference during the dumping of work-pieces,the transported work-pieces are directly unloaded by a loading truck into the grabbing platform of the robot with disorderly directions under the condition of non-manual loading and unloading,and most of them rotated or blocked.In this case,the work-pieces are placed in a naturally stacked state.The positioning of naturally stacked work-pieces is important for reliable transmission and sorting of engineering site materials,and for production as well.In this paper,the 3D reconstruction and pose information output of covered and uncovered target objects are studied on the condition of targets(work-pieces)covered.The three-dimensional identification algorithm is used to determine the outline of the target work-pieces and output the reference point pose information.All the work of researching are listed as follows:(1)Not only two kinds of photography geometry model included in Pinhole Imaging Model and Camera Imaging Model were researched,but also the distortion of the camera is considered in processing of imaging.By using the Zhang Zhengyo u calibration method in binocular stereo vision model,the internal parameters and external parameters of the binocular camera as well as the rotation matrix and translation vectors between two cameras were obtained.(2)A software framework for reconstruction of real-time three-dimensional objects was designed through combining the binocular camera with laser speckle.Because of poor texture information,incomplete surface expression,loss of details,etc.,that generating frequently in three-dimensional reconstructed objects,we combined the matching algorithm based on Bayesian theory with Delaunay triangulation to find out the relationship between support and non-support points,and to complete three-dimensional reconstruction of target objects from sparse matching to dense matching.The reconstruction accuracy of point clouds is improved with less reconstruction losing.(3)In the process of 3D object recognition,a processing method based on the combination of offline training with online recognition,global features and local features is put forward as a result of partial occlusion of the target object.The effective 3D recognition of targets partially occluded or partially lost is realized,and the position and orientation information of target artifact is output by using a voting method.