Research On Parts Detection And Robot Grabbing Based On Improved SSD Algorithm

Under the powerful era of Made in China 2025,intelligent robots are gradually applied to various fields,greatly improving production efficiency and product quality.At present,machine vision is still an important technology that restricts industrial robots.In the current domestic automobile recycling factories,the sorting of small parts,the problem of light and darkness in the factory,and the change of camera pose are still problems for traditional machine vision methods.Challenging problems,such as false detection,missed detection,low recognition rate,and slow detection speed,often affect the sorting of industrial robots.Aiming at the problem that the auto parts recycling factory has poor parts detection results in inaccurate parts detection under actual complicated working conditions,which cannot complete accurate grasping and affect production efficiency,this paper proposes a robotic grasping method based on the deep learning recognition method based on the improved SSD framework.Picking system can realize parts detection,classification,positioning and grasping tasks.First,the target part is detected by the improved SSD model to obtain the position and category information of the part;then the pixel coordinate system is converted to the robot world coordinate system through camera calibration and hand-eye calibration to realize the positioning of the part in the robot space coordinate system;and finally,the robot forward and backward Kinematic modeling and trajectory planning,to complete the task of grasping target parts,the specific work is as follows:1.By comparing the current mainstream target recognition detection algorithms,select SSD algorithms with good open source,real-time and detection accuracy,and then improve the specific network structure according to the detection needs,insert the Inception structure,and optimize its loss function and Non-maximum suppression.After training,the target detection model is compared.After the improvement,the detection effect is improved.After the improvement,the recognition rate is higher.The target detection frame is closer to the real part,which greatly reduces the positioning error.2.Integrated multi-distortion iterative camera calibration based on Zhang's calibration algorithm,and then establish the intermediate coordinate system of the robot end effector to obtain the hand-eye calibration intermediate matrix to obtain the camera external parameters.Through camera calibration and hand-eye calibration,the image pixel coordinates obtained by the target detection algorithm model are converted into the robot world coordinate system,and the conversion between several intermediate coordinate systems is completed.3.Combine the target part detection system and coordinate system conversion to complete the integrated software design,send the space coordinate information of the target part to the robot PC control end,and then perform forward and inverse kinematics modeling and trajectory planning of the NACHI small super-speed robot.The robot uses the coordinate information Make the adaptive adjustment of the corresponding grabbing posture,and reach the designated position to complete the grabbing task.