Detection of overlapped apples in orchard scene using improved K-means and distance least square

Abstract

Automatic detection of mature apples in a complex agricultural condition is still a challenge for an autonomous picking robot due to the influence from overlapping. In order to detecting overlapped apples in tree canopy using a low-cost camera, a robust apples detection and reconstruction approach based on improved K-means and distance least square algorithm was studied. Firstly, the region of potential apple objects was extracted by using improved K-means algorithm. Then, the contours of apples were obtained by utilizing Canny edge detection algorithm on the V component map and the intact contour of unobscured apple was separated from overlapped apples contour after Y-junction searching. Finally, the contour of obscured apple was reconstructed by use of the distance least square circle fitting algorithm. The proposed method was compared with Hough transform method and the experimental result indicated that the proposed method could get much better performance for overlapped apples detection than Hough transform method. Thus it could be concluded that the proposed method is available for robotic apple picking in overlapped fruits scene with low cost.