Control System Design Of Agricultural Picking Robot Based On Binocular Vision

The picking operation of tomatoes is the most time-consuming and laborious process in the whole production process.It is assumed that intelligent picking robot can be used in the process of tomatoes production to realize mechanization,intelligent picking and improve production efficiency.It is not only realistic but also an inevitable trend of agricultural modernization.The accuracy of fruits recognition and location determines the efficiency of the picking robot.According to the characteristics of the picking robot,this thesis has completed the hardware construction of the robot control system,and studied the mathematical modeling and inverse solution algorithm of the picking can robot mechanical arm,as well as the identification and location of the target fruit,realizing the effective identification and three-dimensional space localization of the fruit,providing important information for the autonomous picking operation of the picking robot.The main research work is as below:(1)This thesis introduces the significance of picking robot research and the research status.According to the working characteristics of the picking robot,the hardware of the picking robot control system is constructed.(2)Firstly,this thesis uses the improved D-H method to mathematical modeling of agricultural picking robot manipulator.Secondly,by using coordinate transformation and other related theory to establish a forward kinematics equation.Then the algebra method is used to solve the inverse solution and MATLAB is used to verify the forward and reverse motion.Finally,discusses the five times more and linear mixed with parabolic trajectory planning method.(3)To recognize the single tomato target fruit,firstly,the median filtering is used to deal with the original image,and then the k-means method is used to segment,and performing morphological operation on the image after segmentation.To recognize the overlapping tomato target fruits,firstly,the original image is processed by bilateral filtering,then most kinds of variance method(OTSU)with the improved operator is used to segment the target tomato.After segmentation,the K-means clustering algorithm is used to cluster overlapped target pixel to obtain a single target location.Through the connected domain analysis and region growth,the rough region of a single target boundary is obtained.Finally,the accurate boundary of target fruit is obtained by watershed algorithm.To recognize the green ripe tomato target fruits,firstly,fast normalized cross correlation(FNCC)is used to identify potential areas for green tomatoes and estimate the green target fruit areas.Then Hoff transform is used to identify the location of the fruit.Finally,FNCC is combined with Hoff transform to realize the recognition of green ripe tomato fruits.(4)According to the principle of binocular stereo vision,the internal and external parameters of binocular camera are calibrated by using Zhang Zhengyou calibration method.The feature point matching algorithm based on centroid is given in the paper,according to the matching constraint conditions stereo matching the target fruit.According to the calibration parameters of the camera,the depth value of the target fruit is obtained by using the parallax method,and the three-dimensional orientation of the tomato fruit is achieved.(5)Through multiple picking experiments on the tomato target fruit,the results show that the end-effector is accurate in locating the target fruit and the picking track is steady,which verifies the feasibility of the autonomous picking method and the correctness of the inverse solution algorithm and identification and localization algorithm.