Research On The Technology Of Apple 's Picking Robot Vision System

The fruit production has increased year by year in China, but the shortage of rurallabor force has become increasingly serious. Many research institutions have carried outthe research and development of various agricultural robots. This paper is on the study ofpicking robot vision system. Vision system is the core part of the picking robot. In order tocarry out the accurately operation, the vision system must identify the target precisely.There are many factors to increase the difficulty of visual recognition. For instance, theworking environment of picking robots is natural environment and the job object is thenatural form; the uncertainty of illumination condition and the object size, and so on. Atpresent, there are some studies on apple picking robot in our country, but most of them arestill at the testing stage, which have not been able to enter the orchard to harvest the fruitproducts extensively. The capability of the vision system is the main limit of the practicalapplication of the picking robot.This paper used red apples as the research object. After many analysis of variousedge detection method, this paper used Canny edge detection method to detect the apple’edge. This paper compared the detection results between Canny method and the connecteddomain method, fit one apple image edge to one circle, analyzed the detected error of eachcircle center of the two methods. The average value of error of former method is lowerthan the value of the later method.At the same time, this paper also took the apple trees’ branches as the researchobjects. This paper researched on the color extraction method of branches, and proposed anew method of branches image segmentation. This paper worked out the relationshipbetween R, G, B channels through the accurate calculation method on the color difference.This paper also obtained the threshold segmentation standard, and put forward thecorrection parameters to adjust the threshold rapidly. Through the experiment result, theaccuracy about sharp by extraction methods of this paper could reach80%.Because of the existence of many small holes, burr and impurity in the branch imagesafter segmentation, this paper studies the morphological method to repair the branch images and the apple images. According to the experiment results, the accuracy of thebranch shape extraction can be improved to90%. According to the experiment results, thispaper found that the repair effect of images which the apple is covered at the center of theapple, is better than the repair effect of images which the apple is covered at one side ofthe apple. The cooperation on fruit of the error correction rate, the former method is8%higher than the latter method.In order to calculate the location of branches, this paper studied the image skeletonextraction method of trunk and branches. This paper used the thinning method to extractskeleton of branches. This paper designed the256thinning templates independently. Thispaper divided the type of connection into8categories for analysis, and designed threeexperiments to improve extraction effect. The name of three experiments are regular shapeskeleton extraction experiment, natural shape skeleton extraction experiment and the realtrunk and branches skeleton extraction experiment. According to the experimental results,this paper obtained effective template data to extract image skeleton. This study providesthe two-dimensional information about endpoint location and joint position, which areuseful for three-dimensional position matching to avoid the collision.This paper researched on the camera calibration method and binocular matchingmethod, and obtained the stereo matching scheme. This paper executed errorcompensation according the experimental results. The above research contains fruit centerextraction, partial covered fruit center extraction and shape extraction branches, which areall benefit to improve the recognition accuracy. This paper also has proceeded some basicresearch work about obstacle avoidance. The purpose of these research on this paper is toimprove the accuracy and to ensure the safety of picking robot operation. The researchresults presented in this paper are benefit to further practical application of apple pickingrobot on vision part and provide some reference for other researchers on this field.