A Machine Vision System for the Apple Harvesting Robot

One of the requirements of a fruit harvesting robot is the ability to recognize and locate fruit from the leaf and branch portions; machine vision system is one of the methods used to detect the location of the fruit. This paper showed the development of a machine vision algorithm that would guide an apple harvesting hand prototype. The objectives of this study were to differentiate the fruit from the other portions of the tree, such as the leaf and the branch, and to locate the fruit center and the abscission layer of the fruit’s peduncle. The study was divided into two stages, the recognition stage and the location stage of the apple fruit. In the recognition of the apple, Fuji variety was tested. Apple images were collected using a color CCD camera under natural lighting condition. Color models of Fuji were examined to determine the color properties used to differentiate the fruit from the other portions of the tree, such as the leaf and branch portions. The LCD (Luminance and Color Difference of red) model and the chromaticity model were used for the analysis. The color properties: luminance, color difference of red and chromaticity were tested to determine the thresholds for segmentation using the decision theoretic approach. This approach derived decision functions that could classify the apple fruit, leaf, and branch. The decision functions for the LCD model were dependent on luminance and color difference of red while the decision functions for the chromaticity model were dependent on the trichromatic coefficients, r and g. Segmentation was implemented using multivariable thresholding and the decision functions were the thresholds. Results showed that both models could segment at least 80% of the apple fruits. The location of the fruit center and the abscission layer were determined using a geometrical approach and basic image processing procedures. A geometrical relation between the apple fruit center and the abscission layer was established and combined with standard image processing procedures. The experimental results showed that the method of locating the fruit center and the abscission layer of the fruit peduncle was effective with a success rate above 80%.