Research On Identification And Location Of Automatic Fruit Picking

China's fruit production is high and the national demand is great,but the picking process of the fruit supply chain still relies on manual operation,with high picking cost and low efficiency,unable to meet the supply demand of the market.In order to solve the above problems effectively,using fruit picking robot to replace manual operation will become the trend of modern orchard management.As the core part of fruit picking robot,visual system can effectively improve the recognition and positioning performance of target fruit,improve the efficiency of operation,and reduce the picking cost.In this thesis,spherical fruit in natural environment is taken as the research object,color image and depth image of fruit are obtained based on Kinect machine vision technology,and two-dimensional plane recognition and three-dimensional space positioning of fruit are realized.The main research results and conclusions are as follows:(1)Image acquisition methods are studied,the paper analyzes the features of color space model,R-G color difference method is used to get the fruit for the color figure after the grayscale color,use OTSU automatic threshold segmentation algorithm to extract the target from the image area,fruit for surface close to the background color of the fruit based on gray level co-occurrence matrix is proposed to extract the texture feature of training support vector machine(SVM)texture segmentation method,and using mathematical morphology operation to eliminate the fruit area inside the hole and edge burr,this method has some robustness to illumination.(2)The recognition method of spherical fruits covered by branches,leaves and fruits in natural environment is studied.The coordinates of edge and center of mass are extracted by introducing Empty Matrix,in this paper,a mathematical method is proposed to determine the edge of fruit by using the distance fluctuation curve between the adjacent artificial marking points and the center of Mass on the edge of the target area to obtain the best occlusion threshold,combined with improved clustering hierarchical clustering algorithm to identify the center and radius of the occluded fruits,the spherical fruits with different occlusions caused by different obstacles can be recognized.(3)The calibration process of Kinect camera was studied.The color map and depth mapare registered by using the internal and external parameters obtained by camera imaging model and calibration technology.According to the image pixel coordinate,image coordinate system and the mapping relationship between camera coordinate system for the infrared camera coordinates the fruit surface 3D coordinate of the feature points are calculated respectively by means of least square method of discrete point cloud data interval sphere fitting of the model parameters,and in each interval using spherical polar coordinates formula at a certain point density point by point interpolation calculation,pinpoint the fruit space coordinate of the center of mass of picking robot manipulator grasping motion provide a certain amount of data to support.