Recognition and localization methods of occluded apples based on convex hull theory

Aimed to realize the recognition and localization of occluded apples by branches, leaves and overlapped apples, a method for recognizing occluded apples based on convex hull theory was presented. In the first step, the image was transformed from RGB color space to L*a*b* color space, and K-means clustering algorithm was used to segment the input image to three different categories including leaves, stems and apples. In the second step, mathematical morphology was carried out to get the contours of apples. In the third step, the convex hull of object boundary was used to extract the smooth and real contours. In the last step, the centers and radiuses of the extracted contours were estimated, which were used to localize the occluded apples. In order to validate the performance of the algorithm presented in this study, a comparative test was conducted using Circle Hough Transform (CHT), and the positioning errors were calculated. The experimental results showed that the average positioning error of the method presented in this study was 4.28%, while that of the CHT method was 16.30%. The method significantly improves the accuracy of target positioning and is feasible and effective to recognize occlude apples.