Hand-eye System Design And Close-shot Stem Recognition For Robotic Harvesting Of Trellis Grape Clusters Based On Depth Sensing

At present,the horizontal trellis has widely become a typical viticulture for grapes.In view of the mechanized harvesting of fresh-food grapes under the horizontal trellis,based on close-shot recognition,a hand-eye system based on depth vision was designed to be “eye-under-finger” in this paper.Furthermore,an algorithm and process for the recognition and location of the picking points of grape stems on the close-shot depth vision were proposed and tested.In addition,a farand-close shot strategy was proposed and tested to recognize and locate the grape objects to realize the self-locking of the target grapes.The main research contents are as follows:(1)Depend on the investigation of the growth characteristics and physical properties of the horizontal trellis grapes,the end-effector of the picking robot was designed and developed,and an end-eye system of "eye-under-finger" was developed.This end structure consisting of the endeffector and the hand-eye system is relatively simple and small in size,and can meet the requirements of the end of the picking robot used for horizontal trellis grapes.(2)Taking the depth image of the grape as the research object,according to the depth information feature,the close-shot detection range suitable for the recognition and location of horizontal trellis grape is determined.After the interference information being removed based on the depth data,from the characteristics of depth difference,the method of locating the dividing points between grape cluster and grape stem is proposed.The position of the picking point on the stem is determined by the dividing point,and the spatial coordinates of the picking point can be obtained by combining the depth information.The method only relies on a single depth information to recognize and locate the picking points,and effectively exploits the advantages of depth vision for low light sensitivity.(3)In order to verify the validity and versatility of the close-shot recognition and location,the designed algorithm was applied on the different types of grapes which were developed under the horizontal trellis.The results show that algorithm has a success rate of 88%,80%,80% and100% for the picking points of "bailusha","qiuhong","hongyidali" and "hongdiqiu".And the time is 0.3 seconds,0.28 seconds,0.29 seconds and 0.28 seconds,respectively,so the algorithm's recognition success rate for picking points is about 87.5%,and the average time is about 0.29 seconds.(4)In order to enable the picking robot to quickly identify and locate the target grape under the unknown environment,according to the grapes growth characteristics under the horizontal trellis,a far-and-close shot recognition strategy based on the lowest point of the grape cluster was proposed.The key points in the far-shot and close-shot gestures are determined.The strategy was verified to be effective and reliable according to the results of the indoor experiments.In this paper,under the premise of close-range recognition of horizontal trellis grapes,the configuration of “eye-under-finger” is designed.According to the depth difference feature of depth image,the picking points on the grape stem were studied and identified,and the identification and positioning of different grapes under the horizontal trellis were realized.It has been verified that this method is versatile and rapid for the recognition and location of grapes under horizontal trellis.Experiments are conducted indoor to verify the feasibility of strategy which combined the far-shot and close-shot to realize the recognition and location.