Design And Development Of The Vision System For Round Fruits Harvesting Robot

To solve the practical problems in agricultural picking industry, it is urgent to study the fruit and vegetable picking robot. Due to the complex picking environment and uneven picking object distribution, the research, especially robot vision system research, turns to be much difficult. How to identify ripe fruit quickly and effectively and to find out accurate location of the fruit has became a hot issue.This paper studies a method of identifying and locating globular fruits with binocular stereo vision technology. Its main purpose is to improve identification efficiency and location precision, which paves the way for the research of the entire picking robots. The main contributions are as follows:This paper proposes a new recognition method of fruit - two step method. Firstly, remove all of gray which is the most difficult to remove in the image. Secondly, extract fruit from the image by color component relationship. It also proposes to use Chord mid-point Hough transform algorithm to determine the coordinate center and radius of a circle, and then extract required feature information for locating spatial orientation of fruit. Based on this method, a recognition software which can recognize several fruit colors at the same time is designed. In the end, the recognition effect of this method is proved by processing experiment on 300 images of fruit.By summarizing several camera models and camera calibration methods, a suitable calibration method for fruit picking robot is determined. And then a binocular calibration experiment is carried out using the calibration toolbox of Matlab. Stereo matching algorithms based on characteristic and supplemented by extremely line constraint are used to realize matching problem about the same fruit in two pictures. To acquire 3-D space position in the world reference, triangle ranging principle is used. Calibration experiment results show that the errors can be limited within 1.91 mm while the measuring distance is from 400 mm to 900 mm and the base line is 200 mm.The picking experiment is carried out in the laboratory. First, the recognition software is improved and 3d coordinate calculation function of fruit is added. And then the software is applied to recognize and locate fruit. Visual information is given to navigate mobile picking robot and control the robot to pick fruit. The rate of picking fruit successfully in this experiment is 95%. The result proves the way in this paper suitable for vision system of the picking robot.