3064peduncle detection of sweet pepper for autonomous crop bined colour and 3d检测甜辣椒自动裁剪颜色信息.pdf

IEEERoboticsandAutomationLetters(RA-L)paperpresentedatthe

2017IEEEInternationalConferenceonRoboticsandAutomation(ICRA)

Singapore,May29-June3,2017

PeduncleDetectionofSweetPepperforAutonomousCropHarvesting-

CombinedColourand3DInformation

1∗222222

InkyuSa,ChrisLehnert,AndrewEnglish,ChrisMcCool,FerasDayoub,BenUpcroft,andTristanPerez

—Thispaperpresentsa3Dvisualdetectionmethod

forthechallengingtaskofdetectingpedunclesofsweetpeppers

(Capsicumannuum)inthefield.Cuttingthepedunclecleanly

isoneofthemostdifficultstagesoftheharvestingprocess,

wherethepeduncleisthepartofthecropthatattachesitto

themainstemoftheplant.Accuratepeduncledetectionin3D

spaceisthereforeavitalstepinreliableautonomousharvesting

ofsweetpeppers,asthiscanleadtoprecisecuttingwhile

avoidingdamagetothesurroundingplant.Thispapermakes

useofbothcolourandgeometryinformationacquiredfroman

RGB-Dsensorandutilisesasupervised-learningapproachfor

thepeduncledetectiontask.Theperformanceoftheproposed

methodisdemonstratedandevaluatedusingqualitativeand

quantitativeresults(theArea-Under-the-Curve(AUC)ofthe

detectionprecision-recallcurve).WeareabletoachieveanAUC

of0.71forpeduncledetectiononfield-grownsweetpeppers.

Fig.1.Sweetpepperpickinginoperationshowingaroboticsarmequipped

Wereleaseasetofmanuallyannotated3Dsweetpepperand

withanend-effectortooltoh