Research On Vision System Of Apple Picking Robot Based On Multi-feature Fusion

With the rapid development of robot and machine vision technology,the functions of picking robot are gradually strengthened,which puts forward higher requirements for apple identification and positioning.Orchard environment is complex,fruit posture is different,overlapping occlusion is a difficult problem to solve the identification and location of apple.To solve the above problems,this paper proposes a multi-feature fusion identification and positioning method for apple picking robot.The main contents are as follows:Aiming at the target recognition and quantitative positioning in complex orchard environment,a stereo vision system of apple picking robot was designed.The internal and external parameter matrices of the camera were solved based on Zhang’s calibration method.The spatial pose analysis and experimental research were carried out for binocular camera,and the influence of the number of calibration images on the camera calibration accuracy was verified.Apple images under different environment for the target detection,this paper proposes a using a variety of apple gray feature extraction algorithm of color space,into the variety characteristics,characteristics of gray scale figure K-means clustering processing,consider extracting effect and efficiency determines the apple color features required gray-scale extraction algorithm and the target of the coarse segmentation algorithm.Joint contour second moment method,centroid outline the shortest distance method and mass center contour of variance is used to identify the position of the binary search method to target for apple,the outline for the second order moment method in lower accuracy,put forward the outline of the second moment invariant moment method,to preserve the most can reflect the characteristics of the target attitude,through multiple feature fusion strategies improve the accuracy of the single fruit recognition.Aiming at the low recognition rate of overlapping apples in unstructured environment,an improved region growth algorithm was proposed to achieve fine segmentation of overlapping apples.Based on the statistical model of apple distribution,the feature information of overlapping fruits was obtained,and the depth feature in the apple image and the improved region growth algorithm were used to achieve the fine segmentation of overlapping fruits,which could segment any overlapping apple region of any size.In order to verify the effectiveness of the vision system of apple picking robot based on multi-feature fusion,this paper takes a variety of apples as experimental objects and uses the improved region growth algorithm to accurately recognize and locate apple targets,providing a good technical path for intelligent picking robot.