农业工程学报
農業工程學報
농업공정학보
2015年
5期
167-174
,共8页
王丹丹%徐越%宋怀波%何东健
王丹丹%徐越%宋懷波%何東健
왕단단%서월%송부파%하동건
机器人%算法%水果%苹果目标%目标定位%平滑轮廓%转动惯量%对称轴
機器人%算法%水果%蘋果目標%目標定位%平滑輪廓%轉動慣量%對稱軸
궤기인%산법%수과%평과목표%목표정위%평활륜곽%전동관량%대칭축
robots%algorithms%fruits%apple target%localization%smoothing contour%moment of inertia%symmetry axis
果实采摘点的精确定位是采摘机器人必须解决的关键问题。鉴于苹果目标具有良好对称性的特点,利用转动惯量所具有的平移、旋转不变性及其在对称轴方向取得极值的特性,提出了一种基于轮廓对称轴法的苹果目标采摘点定位方法。为了解决分割后苹果目标边缘不够平滑而导致定位精度偏低的问题,提出了一种苹果目标轮廓平滑方法。为了验证算法的有效性,对随机选取的20幅无遮挡的单果苹果图像分别利用轮廓平滑和未进行轮廓平滑的算法进行试验,试验结果表明,未进行轮廓平滑算法的平均定位误差为20.678°,而轮廓平滑后算法平均定位误差为4.542°,比未进行轮廓平滑算法平均定位误差降低了78.035%,未进行轮廓平滑算法的平均运行时间为10.2 ms,而轮廓平滑后算法的平均运行时间为7.5 ms,比未进行轮廓平滑算法平均运行时间降低了25.839%,表明平滑轮廓算法可以提高定位精度和运算效率。利用平滑轮廓对称轴算法可以较好地找到苹果目标的对称轴并实现采摘点定位,表明将该方法应用于苹果目标的对称轴提取及采摘点定位是可行的。
果實採摘點的精確定位是採摘機器人必鬚解決的關鍵問題。鑒于蘋果目標具有良好對稱性的特點,利用轉動慣量所具有的平移、鏇轉不變性及其在對稱軸方嚮取得極值的特性,提齣瞭一種基于輪廓對稱軸法的蘋果目標採摘點定位方法。為瞭解決分割後蘋果目標邊緣不夠平滑而導緻定位精度偏低的問題,提齣瞭一種蘋果目標輪廓平滑方法。為瞭驗證算法的有效性,對隨機選取的20幅無遮擋的單果蘋果圖像分彆利用輪廓平滑和未進行輪廓平滑的算法進行試驗,試驗結果錶明,未進行輪廓平滑算法的平均定位誤差為20.678°,而輪廓平滑後算法平均定位誤差為4.542°,比未進行輪廓平滑算法平均定位誤差降低瞭78.035%,未進行輪廓平滑算法的平均運行時間為10.2 ms,而輪廓平滑後算法的平均運行時間為7.5 ms,比未進行輪廓平滑算法平均運行時間降低瞭25.839%,錶明平滑輪廓算法可以提高定位精度和運算效率。利用平滑輪廓對稱軸算法可以較好地找到蘋果目標的對稱軸併實現採摘點定位,錶明將該方法應用于蘋果目標的對稱軸提取及採摘點定位是可行的。
과실채적점적정학정위시채적궤기인필수해결적관건문제。감우평과목표구유량호대칭성적특점,이용전동관량소구유적평이、선전불변성급기재대칭축방향취득겁치적특성,제출료일충기우륜곽대칭축법적평과목표채적점정위방법。위료해결분할후평과목표변연불구평활이도치정위정도편저적문제,제출료일충평과목표륜곽평활방법。위료험증산법적유효성,대수궤선취적20폭무차당적단과평과도상분별이용륜곽평활화미진행륜곽평활적산법진행시험,시험결과표명,미진행륜곽평활산법적평균정위오차위20.678°,이륜곽평활후산법평균정위오차위4.542°,비미진행륜곽평활산법평균정위오차강저료78.035%,미진행륜곽평활산법적평균운행시간위10.2 ms,이륜곽평활후산법적평균운행시간위7.5 ms,비미진행륜곽평활산법평균운행시간강저료25.839%,표명평활륜곽산법가이제고정위정도화운산효솔。이용평활륜곽대칭축산법가이교호지조도평과목표적대칭축병실현채적점정위,표명장해방법응용우평과목표적대칭축제취급채적점정위시가행적。
The localization of picking points of fruits is one of the key problems for picking robots, and it is the first step of implementation of the picking task for picking robots. In view of a good symmetry of apples, and characteristics of shift, rotation invariance, and reaching the extreme values in symmetry axis direction which moment of inertia possesses, a new method based on a contour symmetry axis was proposed to locate the picking point of apples. In order to solve the problem of low localization accuracy which results from the rough edge of apples after segmentation, a method of smoothing contour algorithm was presented. The steps of the algorithm were as follow, first, the image was transformed from RGB color space intoL*a*b color space, and then K-means color clustering algorithm was used to detect the apple target. The image was processed with amorphological opening operation with a ‘disk’-shaped structural element of radius 5 beforeK-means clustering algorithm so as to ensure the accuracy of theK-means algorithm. Secondly, image pre-processing algorithms were carried out. Hole filling and area threshold algorithms were performed first to remove noise, and then a mathematical morphology operation with a ‘disk’-shaped structural element of radius 10 was conducted to remove big spurs on the contour of apples. Thirdly, the contour of an apple was extracted by processing the pre-processed image with a morphological open operation. The calculate centroid of an apple and the distance between contour points and centroid were calculated, and then the distance curve could be obtained. After that, wavelet decomposition and Spline interpolation algorithms were used to smooth the distance curve, and then the smoothed distance curve was used to rebuild the contour of the apple. The procedures of rebuilding the contour of apples were as follow: 1) Coordinates transformation. In order to make an image coordinates system in accordance with common used coordinates system, coordinates transformation was needed. 2) Translation of the original point of coordinates to simply calculation. 3) Contour points-centroid angle normalization and calculation, which was of great significance to rebulit contour points. 4) Rebuilt contour points using the smoothed distance curve and normalized contour points-centroid angle. After these four steps, the contour of an apple could be obtained. Finally, the contour was used to extract the symmetry axis of an apple by using a moment of inertia algorithm. In order to verify the validity of this algorithm, a test was conducted by using the original algorithm and the presented algorithm with 20 single and unblocked apple images, respectively. The average error of the original algorithm was 20.678°, and the average error of the presented algorithm was 4.542°, 78.035% less than that of the original algorithm. Furthermore, the average run-time of the proposed algorithm was 7.5 ms, which was decreased by 25.839% when compared to the original algorithm (10.2 ms). The results showed that the presented algorithm could locate the picking point of an apple accurately and effectively. In conclusion, the presented algorithm is feasible for extracting the symmetry axis and locating the picking point of apples. However, this method was not applicable to blocked apple images, uneven illumination apple image, images containing apples with poor symmetry, and apples with part of a green region, for the entire contour of apple in these images cannot be obtained.