Design And Development Of Vision And Control System For A Picking Robot

In recent years,with the increase of the income level of Chinese people and the change of dietary philosophy,the consumption of fruit has increased dramatically,and the scale of domestic fruit cultivation has been increasing with the stimulation of market demand.But so far that the production of fruit in China is still mainly by hand,and a large number of manual picking work not only cost money and labor,but also is difficult to guarantee the quality and the appearance of the fruit after picking.It is of great practical significance for the continued development of the fruit industry to carry out the automatic production of fruit and reduce the artificial production link.At the same time,it has gradually developed into a hot research direction in the field of agricultural mechanization.This paper introduces the overall layout of picking robot based on the analysis of the data of existing picking robot and the theory of control theory and machine vision.It use that modular concept to construct the test sample from two parts of binocular vision platform and simulated picking platform in the design.In the research of the functionality,The object recognition based on color feature and the stereo matching based on centroid is completed.And the optimized neural network is used to rack a target in space.Finally,the simulated picking experiment was carried out and the results of simulated picking experiment were studied and analyzed.In the first part,the visual system imaging model was built based on the study of machine vision theory.In the second part,the color feature is used to realize the segmentation of the target.By studying the gray change of different color components,the image segmentation is realized and the recognition of the target fruit is achieved under the 2R-G-B component.In the third part,the stereo matching theory is studied,and the matching method based on the centroid is proposed to realize the matching of the left and right images.In the fourth part,the improved BP neural network algorithm is applied to the camera calibration and target space positioning by studying the BP neural network algorithm.And the transformation from centroid coordinates to three-dimensional coordinates is realized.In the fifth part,the motion function library based on the MAC-3003SSI2 control card is studied in the serial robot platform.Moreover,Visual C++ is used to realize the secondary development of control software,and the planned picking action has been completed.Finally,the feasibility and effectiveness of the research scheme are tested by summarizing and analyzing the simulated picking data.