自然环境下苹果作业机器人双果重叠目标侦测方法

为进一步提升苹果的侦测精度,从而提高苹果作业机器人的果实作业效率,以果园重叠苹果为对象,研究了自然环境下双果重叠目标的机器侦测方法。首先将采集到的苹果图像在Lab色彩空间中利用K-means聚类算法提取重叠苹果目标区域;其次在得到重叠苹果边界上的Harris角点后,通过关键角点检测算法定位苹果重叠部分的果实轮廓所在区域,并利用Canny边缘检测算子提取出苹果重叠部分的果实轮廓;然后利用Y型节点搜索算法实现重叠苹果目标的单果轮廓分离,并得到未遮挡果实的完整轮廓;最终利用距离最小二乘算法对被遮挡苹果目标进行果实轮廓重建。为验证方法的有效性,将Hough变换法、Spline样条内插法的重叠果实侦测结果与本文方法所得结果进行对比。试验结果显示,本文方法不仅可以侦测出未遮挡果实的完整轮廓,同时对被遮挡苹果目标也有较好的轮廓重建效果,其果实平均重合度和定位误差分别为95.43%和4.44%,侦测性能明显优于前两种方法,表明本文方法可以较好实现自然环境下苹果双果重叠目标的侦测,该结果为苹果作业机器人多果重叠目标的自动化侦测提供参考。

Detection of Double Overlapped Fruits in Natural Scene for Apple Operation Robot

In order to improve the precision of apple detection and further increase the work efficiency of apple operation robot, this paper studied the detecting method with double apples overlapping target under natural environment. Firstly, the region of apple objects in image was extracted by using K-means clustering algorithm under Lab colour space transforming from RGB colour space. Secondly, the key corner points on the apple outline were found among the Harris corner points by calculating the extremum of distance curve between every corner point and the centroid of the overlapped fruits. Then the contour of unblocked apple in overlapped area was extracted by utilizing the Canny algorithm in the detecting window located by key corner points. Thirdly, Y-junctions detection algorithm was utilized to separate contour of individual apple from overlapped apples, and then the complete contour of unblocked apple was obtained. Finally, the missing contour of partly blocked apple was reconstructed by using distance least square algorithm. In order to verify the validity of this method, the Hough transformation method and Spline interpolation method were used to compare with the proposed method for overlapping apples detection. The experimental results showed that the proposed method could not only detect the complete contour of unblocked apple, but also preferably reconstruct the missing contour of partly blocked apple. The average coincidence degree and average error of proposed method were 95.43% and 4.44%, respectively, and its detecting performance was better than that of the former 2 methods, indicating that the proposed method could better detecting double apples overlapping targets under natural environment. Above results provided reference for automatic detection of apple operation robot with multiple fruits overlapping targets.