Design The System Of Tomato Recognition And Localization Based On Structured-Light Vision

The automatic harvesting robot for tomato has been researched since 1990s, which aimed at decreasing cost and labor-intensity for picking mature fruits. As one key component of the robot, the vision unit is used to identify and locate the mature fruits, and plays the most important role for the robotic harvesting under the agricultural non-structure condition. However, it is proved as the bottleneck for robot’s commercial employ, because it is greatly difficult to accurately obtain the fruit information in agricultural condition, such as the variable illumination in greenhouse, the complex background behind the fruit and the random distributed and overlapping targets.According to the tomatoe which grows in greenhouse, a new method of identification and localization based on linear structured-light vision system is proposed. It use the locating features of linear structure-light to solve the problem of target matching causing by occlusion and the segmentation problem of overlapping targets.The contents and results of the research can be briefly summarized as follows:(1)According to the requirements of vision system used for tomato harvesting robot, a linear structured-light vision system is designed. Based on the camera imaging geometric model and linear structure-light model, the localization principle of this vision system is analyzed. Then the camera internal parameters and the linear structure-light parameters are determined by calibrating.(2)The process of recognition and localization based on linear structured-light vision is introduced. The |R-G| chromatic aberration image of the tomato is proved to be the most available to the segmentation after analysing various kinds of extraction algorithm for the mature tomato, especially under the condition of different light. A method combines the direction for image of light stripe and gray barycenter method is used to extract the center of laser stripe area after analysing various kinds of methods. The locating method is analyzed. For a single fruit, take the fitting circle’s center as the fruit’s center. For overlapping ones, firstly fill the center line with gray which maps the depth information of the line, than take out the lines which belong to a single fruit by threshold segmentation, finally get the fruit center through space sphere fitting.(3)The whole tomato harvesting robot is systematically designed, including rail cars, mechanical arm, picking end executor and line structured light vision system. And according to the requirements of tomato harvesting robot and work flow of the tomato harvesting robot, a control system of tomato harvesting robot is developed based on the Windows developent environment and Visual c++ programming language.(4)As the field test results show, the positioning error of the linear structure-light vision system is less than 10mm and the success rate of harvesting robot is above 80%.