Research On Robotic Grasping Region Localization Through RGB-D Camera

With the widespread use of depth cameras,RGB-D data based applications have grown rapidly in each passing day such as in indoor scene modeling and automatic navigation.Meanwhile,price of depth camera is also becoming more and more cheap.However,RGB-D data driven robot grasping still cannot well meet human require-ments due to accuracy and efficiency problems in vision systems.In this paper,we present a novel framework to drive automatic robotic grasp by matching camera captured RGB-D data with 3D meshes,on which prior knowledge for grasp is pre-defined for each object type.And then propose a novel grasping lo-calization algorithm based on RGB-D data to drive robot.This proposed framework is composed of four parts:region segmentation,object classification,points cloud regis-tration and robotic driver.This algorithm framework can be summarized in the following sections:1.Build 3D model and mark grasping region:scan 3D meshes for typical object shapes and pre-define grasping regions for each 3D shape surface,which will be considered as the prior knowledge for guiding automatic robotic grasp.2.Filter object regions based on RGB-D:segment RGB-D image captured by a depth camera without semantics from background using depth data,and then recognize 2D shape of the object by a SVM classifier.3.Points cloud registration:propose a new algorithm to register the segmented points cloud data generated by depth camera inherent parameter with predefined 3D meshes to gain transformation matrix,and finally convert points cloud data of grasping region to camera coordinate.4.Robotic drive grasping:estimate the external matrix with chessboard calibration and reflect camera coordinate to robotic coordinate.Our experimental results show that the proposed framework is particularly use-ful to guide camera based robotic grasp.At the end of the paper,we also give the demonstration of our automatic grasping region localization system.