Study On Picking Robots For Ediable Rose

Starting from the image identification of the roses and in combination with the growth characteristics of the roses, this paper researches the method for the rose image identification. Under the RGB color space, the line section plan analysis is conducted to the image, to establish the split pattern based on the ratio between G and R; by utilizing the LRCD difference transformation method, realize the split of the roses and the background; and the optimal threshold of H component and S component under the HIS color space. Considering the influence of the light ray on the image and comparing the influence of the light ray o the split effect, it is found that the split effect based on S component under the HIS color space is optimal. Studying the feature extraction based on the image content, it is found that the distinction degree of the textural features of the roses is relatively great with the rotation invariance. The identification model is established by combining the BP neural network and is used to identify the roses in the flower bud and picking period. The test results show that:(1) the threshold segmentation based on the S component can avoid the illumination influence well, in the302roses,301roses are successfully split, with the accuracy over99.6%;(2) the comprehensive identification rate of the identification model of the roses in the picking period based on the textural features and the BP neural network is greater than85%, and is not sensitive to the light ray but is greatly influenced by the training sample.Stimulating people to pick the roses, the arms of the picking machine are determined to have four degree of freedom, including the waist rotary joint, the big arm rotary joint, and double the small arm rotary joint; in accordance with the cultivation method, the structure parameters of each joint are determined; in accordance with D-H homogeneous variation method, the kinematical equation is established and the inverse analysis of the kinematical equation and the Jacobian matrix are provided; the stimulation of the picking arms is realized by utilizing ADAMS, thus the data from stimulation correspond to the solution of the kinematical equation; the structure of the end effector is determined in accordance with the biological characteristics of the roses, including:the grasping fingers, the sleeve and the cylinder, the scape split is realized via the pneumatic blade in the cylinder, in combination with the experiment data, the net intensity of pressure in the cylinder at least reaches7.4kPa.The monocular vision grabbing model is developed on the basis of pinhole imaging principle. The basic idea:the pixel of two-dimensional flat image corresponds to the straight line of three-dimensional space, if mechanical arm can move along the straight line, it can be sure to get close to the target point; two-dimensional image can only give directions of the space target, if we want to locate the space target, the grabbing depth must be known, so we combine with the proximity sensor to detect the target. We achieve the track simulation of the monocular vision model utilizing MATLAB. In consideration of there is error between the monocular vision model and the rotary joint of the mechanical arm, we introduce error into the simulation. We discover from the trajectory chart that because of the error, the mechanical arm can move around the space line fluctuation and get close to the target point.