Research On The Method Of Generating Visual Navigation Path Of Kiwi Picking Robot

This paper mainly studies the visual navigation path generation method of kiwi picking robot,the kiwi fruit orchard of the kiwi kindergarten in Meixian County of Shaanxi as a research object.Analysis of the operating path of kiwi fruit and kiwi picking robots in the scaffolding cultivation mode,the final decision to use monocular vision of the navigation,Image of kiwi fruit orchard was collected by CCD color camera.The image information is processed by MATLABR2012 a and extracted the navigation path information.The relative pose parameters of the kiwi picking robot are solved by the calibration of the camera and the information of the navigation robot.Finally,the lateral deviation and heading deviation of the kiwi picking robot relative navigation path are obtained.In this paper,a navigation path generation algorithm for kiwi picking robot suitable for scaffolding cultivation mode is proposed,and the algorithm validation experiment is carried out on the platform of the picking robot.The main contents and achievements of this paper are as follows:1.A method of kiwi picking robot navigation based on machine vision is proposed: based on the cultivation mode of kiwifruit orchard,proposed the intermittent walking project for kiwi picking robot.Through the investigation and study,the orchard line width 4m,plant spacing 3m,kiwi fruit trees 1.8m high.According to the distribution of the growth position of kiwi fruit,the optimal path of the kiwi picking robot is to walk along the edge of the kiwifruit fruit line.The boundary information of the tree line is collected by the monocular vision system to construct the kiwi picking robot navigation visual model.The camera has a height of 1.2m and a 15 ° down.After the collected images are processed,the least squares method is used to fit the line boundary line,and the relative angle and distance information of the kiwi picking robot relative to the tree boundary are obtained.Which lays the foundation for the path planning and tracking control of kiwi picking robot.2.The calibration method of the navigation camera is studied and the camera is calibrated: combined with Zhang Zhengyou camera calibration method,use MATLAB camera calibration toolbox to calibrate the navigation camera.Get the inside and outside parameters of the navigation camera,xf = 1680.26995(pixel),yf = 1671.50437(pixel),the image center point coordinates(523.0,398.5).And finally establish the relationship between the world coordinate system,camera coordinate system and image coordinate system.3.Study on the image pretreatment and image segmentation of Kiwi Fruit orchard image: The monocular vision system of kiwi picking robot was constructed,the development of monocular vision system based on PC platform was carried out.Using MATLAB R2012 a software to write image processing program.This paper studied the identification of the boundary of the fruit trees by the kiwi picking robot visual navigation system,and treated the kiwi fruit orchard images with the ridge and the trunk as the recognition object respectively.The process of identification of orchard images is: color space selection,filtering,image enhancement,threshold segmentation,noise removal.Using the H channel image to extract the ridge feature,the threshold value of the H channel is(28)4745.0T.Using the method of homomorphic filtering and regional growth to effectively extract the trunk target,kiwifruit trunk area average gray value is 3.81,the threshold threshold is ?30T(28).Separate two identifiable targets and prepare for the extraction of navigation parameters.4.This paper proposes a method to extract the characteristics of kiwifruit tree and the target of the ridge feature and the path generation method corresponding by the two objectives: In this paper,the kiwi picking robot vision navigation is mainly to extract the edge of kiwi fruit line,extract the edge of the feature point,according to the two characteristics of the target design of different feature points of the extraction algorithm.Taking the ridge as the characteristic goal,scanning the segmented image to select the typical feature target point;taking the trunk as the characteristic goal,marking the trunk feature with the smallest rectangular box,extracting the intersection of the trunk and the ground;The least squares method is used to fit the navigation path.The path model is y = kx + b.According to the information of the navigation path fitted in the image,the relative angle and distance information of the kiwi picking robot relative to the tree boundary are obtained.And the generated navigation parameters were tested statically.The average relative deviation between the calculated navigation baseline and the navigation point of the artificial fitting navigation baseline is 1.28%.The simulation experiment is carried out in the MATLAB environment and analyzed the results.5.To build a kiwi picking robot mobile platform control system to verify the test: there are two rows of trees on the campus of the North Campus of Northwest A & F University.The experimental environment are similar with the scaffolding cultivation mode "Hayward" Kiwi orchard.The results showed that the pathogenesis of kiwifruit orchard was 25 m,and the average deviation was 11.88 cm.The results showed that the path generation method proposed in this paper could provide theoretical basis and technical support for the walking of kiwifruit picking robots.