In apple harvesting robot, the first key part is the machine vision system. Identifying single objects from fruit images is the first and foremost task in machine vision system. However, the main problem affecting the identification of single fruits is that fruit regions in image taken in unstructured orchard environment are overlapping in some cases. On the basis of studies on fruit image segmentation and boundary tracking, this paper proposed a new method for partition and recognition of overlapping apples based on boundary curvature. After the whole one-pixel boundary of overlapping fruits was extracted, the curvature at every point on the boundary was calculated. Then the whole boundary was split into segments by removing points with abrupt change in curvature and valid segments were retained through a screening process based on three criterion. At last, the identification of fruits was achieved by circle fitting and merging. Ultimately, the experimental results show that the proposed method is effective.
[1]
Chen Zao-sheng.
Region labeling based on run-length coding for binary image
,
2004
.
[2]
Laurent Wendling,et al.
A new method to detect arcs and segments from curvature profiles
,
2006,
18th International Conference on Pattern Recognition (ICPR'06).
[3]
Giorgio Grasso,et al.
Localization of spherical fruits for robotic harvesting
,
2001,
Machine Vision and Applications.
[4]
N. Otsu.
A threshold selection method from gray level histograms
,
1979
.
[5]
Gao Rui,et al.
Object extraction for the vision system of fruit picking robot
,
2004
.
[6]
Fan Jun,et al.
Recognition of mature oranges in natural scene based on machine vision
,
2008
.
[7]
Weitong Zhang,et al.
Recognition and Location of Fruit Objects Based on Machine Vision
,
2006,
ICAT.