Research On Blade Recognition Robot Based On Machine Vision

With the continuous improvement and development of science and technology,the trend of global warming is obvious.The carbon sequestration of plants has drawn more and more attention.Only by increasing the capacity of the environment to absorb and store carbon can it reduce the concentration of carbon dioxide in the air and thereby inhibit the deterioration of the greenhouse effect.Carbon sink measurement method has also been the subject of research,compared to the traditional macro-statistical methods,micro-measurement more and more people recognized.At present,the method of microcosmic carbon sink measurement relies on the surveyor holding a photosynthesis measuring instrument to clamp the local blade.However,every hour,the surveyor needs to manually replace the clamping blade,which is a waste of human and material resources.Therefore,manual measurement instruments should be replaced by automatic measurement instruments.This paper aims at the automatic clamping of the leaves of green plants,the Kinect vision sensor and the TurtleBot_Arm manipulator.Some of the issues involved in smart clamping include the detection of potted plants based on deep learning,point cloud segmentation and pose calculation of plant leaves,hand-eye calibration and motion planning for four-degree-of-freedom robotic arms.In this paper,the research was carried out respectively.Finally.a set of completed hardware and software systems for intelligent blade clamping was built.The research mainly includes the following aspects:(1)In order to achieve the recognition of potted plants,this paper by comparing and analyzing the deep learning target detection framework,the Coco data set and Resnet pre-training model were used to carry out the migration training of the Faster-Rcnn model.Under the verification of the Coco data set,the identification accuracy mAP can reach 0.37.The computation time can reach 0.6s under the CUDA parallel acceleration of the Nvidia Gtx1070 graphics card,which can basically satisfy the real-time blade recognition and clamping.(2)In order to achieve the posture of the blade extraction,this paper by using the coordinates of the potted plants in the image,selecting the color image and the depth image in the coordinate area can generate a color point cloud.The colored point cloud was subjected to cumulative sampling,color separation,and Euclidean clustering to obtain each target blade point cloud.Then,the OBB bounding box algorithm was used to estimate the position and orientation of the blade center point for robotic arm clamping.(3)In order to get the coordinate transformation relationship between the camera and the robot arm,this paper recorded 10 sets of robot end-effector poses and Aruco_Marker poses and by solving the X-component solution of the matrix equation AX=XB,automatic hand-eye calibration of the robot arm was achieved.(4)In order to achieve the collision avoidance planning of the robot arm,this paper by using the ROS-Moveit advanced motion planning,3D perception,dynamics,and control and navigation platform to complete the collision avoidance path planning and clamp control of the end effector of the robot.In this paper,the software and hardware system of the automatic blade grab robot is constructed by using the potted plant identification,blade position and posture extraction,hand-eye calibration,and manipulator motion planning,and a large number of tests and verifications are carried out in the actual clamping process.The experiment basically achieved automatic clamping under unsupervised conditions of some plants,providing a reliable way of thinking for carbon sink measurement automation.