关键词: 智能采摘, 重叠分割, 轮廓重建, 曲率, 距离分析

Abstract: Accurate identification of overlapped fruits in natural environment is one of the problems that need to be solved in intelligent picking. In this paper, a new method of segmentation for mature overlapping citrus in the natural environment was proposed based on contour curvature and distance analysis. Firstly, the contours of overlapping citrus fruits were extracted and Gaussian smoothing was performed. The abnormal contour pixel points were found by means of the curvature analysis. Secondly, the two adjacent abnormal pixel points were sequentially connected to form a line segment, and the distance from the pixel point to the contour on the line segment was analyzed. The overlapping citrus contour segmentation was completed at the intersection of two adjacent normal segments, and the corresponding non-citrus contour pixels were eliminated by finding the abnormal line segment. Finally, the obtained citrus target was reconstructed by the least squares ellipse fitting method. The experimental results showed that the average error, non-coincidence and time of the citrus targets obtained by this method were 4.903%, 5.593% and 0.408 s, respectively, which were better than those of Hough transform algorithm and RANSAC algorithm, and could meet the intelligent identification requirements of mature overlapping fruits in natural environment.

Key words: intelligent picking, overlapped segmentation, contour reconstruction, curvature, distance analysis