Research On Fruit Recognition And Locating System Of Citrus Fruit Based On Human Eye-like Platform

The application of machine vision technology to fruit and vegetable picking robots improves the environment perception ability of the picking robots and realizes the automation of fruit and vegetable picking,which is of great significance to the intelligent development of agricultural machinery.Based on the human-like active vision platform,this paper develops a citrus picking robot fruit recognition and positioning system.The work of this article is mainly divided into the following aspects:(1)By studying the motion form of the human eye,develop a Human-like eye active vision platform to improve the ability of the picking robot to obtain a wide field of view.First,the overall design of the mechanism was determined according to the subject requirements,and a 4-DOF human eye-like platform based on the series mechanism was designed.Then use the DH parameter method to model the kinematics of the human-like platform,and verify the correctness of the DH modeling parameters on the MATLAB software.Secondly,use the ADAMS software to verify the extreme motion of the selected joint motor torque.The simulation results prove The selected motor meets the working requirements.Finally,build the hardware control system of the human eye platform;(2)The segmentation and recognition algorithm of citrus image is researched,and the image pixel coordinates of citrus center are extracted.First of all,the grayscale processing of citrus image is studied,combined with the existing grayscale factors,an improved R-G color difference grayscale factor is proposed,which can well separate the citrus from the background.Then,on the basis of the R-G grayscale factor proposed in this paper,combined with the Otsu method,the citrus image is binarized,and the segmented image is morphologically processed to eliminate image holes and reduce interference.Then,the edge extraction algorithm is determined by comparing the image processing results of different edge operators.Taking into account the shape characteristics of the citrus fruit,the Hough circle transform algorithm is used to fit the contour of the citrus and return the central coordinate value,which provides a preliminary basis for the realization of binocular ranging.(3)A Human-like eye vision system model was established,and the three-dimensional positioning method of citrus fruit was studied.Firstly,the principle of camera imaging and calibration was introduced.Aiming at the problem that the external parameter matrix of the vision platform needs to be recalibrated when the pose changes,a method of estimating the external parameter matrix based on joint angle was proposed,and the principle of 3D reconstruction was introduced.Secondly,the platform vision system parameter calibration experiment was carried out.For the internal parameter matrix,Zhang’s calibration method was used to calibrate the result and for the external parameter matrix,the external parameter matrix estimation experiment was carried out by the proposed method,which verified the effectiveness of the external parameter calibration method of human eye platform.Finally,SURF algorithm was used to generate image feature vector,and SSD cost matching function was used for bidirectional matching of binocular orange image.At the same time,RANSAC algorithm was used to eliminate the mismatching points.The matching of binocular corresponding citrus and 3D coordinate calculation of picking point were successfully realized.(4)The host computer of the human-like vision system is built,and the software construction process and functions are explained.The citrus positioning test was carried out on the human-like eye platform.The research results showed the effectiveness of the motion performance of the human-like eye platform,and it can identify and locate the three-dimensional coordinates of citrus fruits,which greatly improves the intelligent level of the citrus picking robot.