Research Of Vision-Guided Robot Bin-Picking System For Air Switch Part

In recent years,with the advent of "Made in China 2025" program,industrial robots are becoming more and more popular in the manufacturing industry.Thanks to the continuous advancement of sensor technology and numerical control technology,the application of industrial robots is developing vigorously and gradually towards intelligence.In the development of intellectualization,adding visual sensors for industrial robots is an effective means to enable robots to obtain information and feedback from the surrounding environment and greatly improve the automation level of robots.Nowadays,vision-guided robots have become a new direction of common concern between robots and computer vision.In the field of automatic assembly of vision-guided industrial robots,most domestic robot application scenarios still need external facilities to co-operate,and can not complete the automatic assembly very well.The main technical bottleneck is parts recognition and location when robot picks up.Aiming at this problem,this thesis develops a scattered stacked parts picking system based on industrial robot and binocular camera.The picking object is the conductive film in the air switch.The main tasks are as follows:First,the actual project requirements are analyzed,and the overall scheme of parts identification and location is studied.Starting from the structural characteristics of the picking parts itself and combining with the posture of the parts placement,the overall picking scheme flow of the stereo vision system based on binocular camera combined with gray code structured light is proposed.The hardware and actuator of the system are selected and designed,and the software architecture is designed.Secondly,3D reconstruction is based on binocular stereo vision.The mathematical model of binocular camera is established,and the calibration method of Zhang Zhengyou is used to calibrate the two cameras.The internal parameters of the camera and the position relationship between the cameras are obtained.According to this,the collected photos are preprocessed,deformed and corrected,and then the gray code structured light is used to match the pixels of the same name,and the principle of triangulation is used to reconstruct the three-dimensional image.The maximum error of the model test is 1.404 mm.Then,according to the pickable area of the parts,the parts stack on the two-dimensional image is segmented based on Mask R-CNN,and the whole point cloud model obtained from the three-dimensional reconstruction is combined to get the point cloud of the pickable area in each parts.Then the point cloud normal is calculated by PCA algorithm,and the ISS operator is used to extract the point cloud key point.Study the geometric primitive fitting algorithm based on SAC-IA algorithm and ICP algorithm for target parts pose registration based on its CAD model.Finally,we use the vision system on industrial robots to verify the experimental results.For scattered parts,this thesis design a strategie that chooses the effective pick up area of partss to recognize,combined with the good robustness characteristics of Mask R-CNN,can directly identify the pile of parts from the main camera.Together with the point cloud data acquired from a binocular stereo vision system in the estimated position information,the whole system can accomplish the robot bin-picking task on scattered parts in a good way.