3-D Object Recognition And Location Based On Machine Vision

Robotic grabbing is the most basic task of industrial robots in daily production operations.In the current development of industrial robots,most of the working scenes are still in the offline teaching,fixed-path pre-determined tasks.This kind of task can only be a structured and singular scene,but with the continuous upgrading of work tasks,the demand for dynamic production scenes is getting larger and larger,and intelligent robot crawling is particularly important.This topic takes the intelligent grasping task as the demand,and the target object identification and positioning in the task is taken as the key problem.The depth sensor is used to construct the 3D crawling environment for the robot,and the deep learning is used as the tool to improve the scene recognition ability of the robot.Starting from the robot vision system,this paper introduces different visual system layout methods under different working scenarios,and selects the visual system construction method outside the eyes for the specific tasks of this subject,and carries out the calibration of the camera in the visual system.In order to improve the ability of the robot to identify and locate objects in complex scenes,an object pose estimation method based on image semantic segmentation technology is proposed.It mainly includes two aspects of object recognition and pose estimation.In the method selection of object recognition,this paper builds a pixel-level classification,namely semantic segmentation,by means of a full convolutional neural network in deep learning.Firstly,the RGB image captured by the RGBD sensor is placed in the semantic segmentation network to complete the segmentation and object classification of the image,and then the segmented target object is registered with the depth map to obtain the target object point cloud map,and the point cloud map and the model library are The model uses the ICP(Iterative Closest Point)algorithm to estimate the pose of the object.The visual system construction method,hand-eye calibration algorithm,image semantic segmentation and object pose estimation,which are studied in this subject,basically cover all the key issues of the visual part of the robot capture task,which can not only help the robot to complete the intelligence more efficiently.Grab tasks,and have a certain reference value for each small field.