Research On Tomato Fruit Growth Pose Identification And Picking Control Based On Binocular Vision

In order to make tomato production easy to manage and improve the yield of tomato,the planting model of tomato is mainly based on greenhouses in our country.The traditional manual method of picking tomatoes has the disadvantages of time-consuming,which is an agricultural production link that needs a large amount of labor in a short time.With the continuous improvement of agricultural mechanization level in China,picking robots are increasingly used in the fields.However,the existing tomato picking robots still have many problems.Firstly,the recognition accuracy of tomato fruit is usually terrible due to the change of light intensity and excessive occlusion between fruits and branches.Secondly,the structure and control system of the end effector are complicated due to the unreasonable design.Besides,the damage to fruits and branches is also a problem that the tomato picking robot has been solving because of roughly selecting the picking point and the unreasonable picking motion planning of the manipulator.Based on the above description,this thesis took the cluster growing tomato in greenhouse as the research object,and proposed a method to obtain the fruit growth posture information based on binocular vision system.A multi-knuckle end-effector was also designed and manufactured.Based on the linear interpolation method,the manipulator motion planning was carried out.In addition,a harvesting robot was developed and built that could automatically harvest tomato fruits,and the content listed below is the main conclusion of this thesis.(1)Tomato images collected in a real greenhouse environment were used as the data source of this study,and the I color component in the YIQ color space was selected as the image segmentation factor.Based on the iterative method,the automatic threshold segmentation of the image was completed.The average number of iterations was 4,and the average optimal threshold was 53.09.The accuracy of tomato fruit recognition based on morphological operation and Hough transform was 93.2%.(2)Based on the moment feature,the center line of the fruit axis was characterized and identified in the image.Through a comparison test with the actual center line area marked,it was proved that the correct rate of the fruit center line recognition was 85.5%.Based on the BM binocular stereo matching algorithm,the theoretical spatial position information of the fruit axis was solved.A fruit axis posture compensation algorithm based on the area of the calyx and the contour area was proposed,and the group experiment proved that the average absolute error of the fruit axis pose recognition was9.7%.A picking strategy based on the fusion of pixel area and depth information was proposed as well.(3)By analyzing the geometric and mechanical properties of ripe tomato fruits,it could be seen that the average size of the horizontal axis was 67.48 mm,and the average size of the vertical axis was 56.39 mm,and when the maximum deformation was 3mm and the maximum extrusion force was 10 N,the peel and flesh of tomato fruits were not damaged.The structure design and motion analysis of the multi-knuckle end-effector were carried out,and the maximum inscribed sphere diameter of the end effector was105.20 mm while minimum was 77.83 mm.The prototype of the end effector was finished,and the pneumatic system of the end effector was designed.Through force analysis and grasping force test,the working pressure of the pneumatic system of the end gripper was determined to be 0.75 Mpa.(4)Based on D-H parameter method,the joint coordinate system of 6-DOF manipulator was established.The inverse kinematics analysis of the manipulator was completed based on the analytical method,and the Jacobian matrix was analyzed and solved.Based on the linear interpolation method,the path planning of the manipulator was completed,and the trajectory and speed of the end effector were obtained by MATLAB simulation.(5)The prototype construction and control system design of the tomato picking robot were completed.Fruits with different growth postures were selected for picking tests.Test results showed that about 83.3% of the total tomato were harvested successfully,and the average deviation of the path planning starting point in the robot base coordinate system was 8.73 mm,6.13 mm and 9.83 mm,the average angular deviation between the center axis of the end effector and the center line of the fruit axis was 18.9°.