著者

庄野 浩資 天羽 弘一 高倉 直

出版者

日本農業気象学会

雑誌

農業気象 (ISSN:00218588)

巻号頁・発行日

vol.45, no.2, pp.87-92, 1989-09-10 (Released:2010-02-25)

参考文献数

7

被引用文献数

1 2

---

A method to detect cucumber fruits in the canopy using image processing technique has been studied, with the aim of robotic harvesting. This method distinguishes a fruit from the other parts of the plant, not by the difference in their colors, but by the difference in shapes.The brightness in the digitized image was differentiated in horizontal direction, and clipped into three values (+, 0, -) by two thresholds. Then ridge-lines were drawn between (+) value area and (-) value area.The ridge-line images ordinarily contain some noise from the other parts of the plant. Accordingly, two images for the same scene were taken, one was illuminated from the left and the other was illuminated from the right. The center line image was obtained by comparing the two ridgeline images in consideration of the shift of the fruit's ridge-line which was calculated by a model. After the noise was reduced in the center line image, the center line of the fruit remained.This method contains four unknown parameters, and it was found that these parameters have the optimum value. But there were some deviations in the optimum values obtained from different images. The deviations were not so severe as to make the detection impossible in our experiment, but in general it is necessary to determine the optimum values applicable to many other scenes.This method was applied to some images taken in the field, and showed that it was able to detect the fruit when it was fully illuminated from both sides. The detection, however, became difficult when the light was intercepted by leaves and the illumination for the fruit was not enough.