Research On Visual Positioning Strategy Of Industrial Robot For Metal Casts

Intelligent manufacturing is the inevitable trend of advanced automation and integration technology.With the development of technology,the advanced manufacturing equipment industry has higher requirements for the intelligence level of industrial robots.Machine vision is the core technology to realize the intelligence of industrial robots,and it is also one of the key research directions of intelligent robots.The visual positioning technology of industrial robot is to measure the position and pose of the target workpiece by machine vision.The processing of metal casts will produce a large amount of dust,and the processing environment is often accompanied by sound and light pollution.There are stains and scratches on the surface of the workpiece,which make the imaging conditions of the visual system poor;There are irregular mirror reflection on the surface of rough-machined workpiece due to the previous processing process,which makes the local features of the workpiece overexposed and underexposed,and easily leads to the instability of visual system sampling information,which restricts the application of industrial robots in cast processing industry.Aiming at the problems existing in the measurement of grasping pose of metal casts by industrial robot vision system in unstructured environment,the research on the industrial robot vision positioning method for metal casts processing is carried out to improve the target recognition rate and positioning accuracy of the industrial robot vision system for metal casts.The main research work is as follows:(1)Aiming at the problem that the positioning accuracy of the visual system is not high due to the interference of ambient light and rough workpiece itself,such as rust,burr and scratch,in the 2D visual positioning of rough metal casts,the VoV-YOLO visual positioning method for rough casts is studied to effectively suppress the adverse impact of ambient light and rough castings on the positioning accuracy of the visual system.VoVNet and YOLOv4 network are combined,and a VoV-YOLO target detection network with both advantages is proposed to locate the local features of the target workpiece stably and quickly.On this basis,the modified template matching algorithm(De-linemod,Deformable Line-MOD)of deformable template technology is used to locate the local features of rough castings with high precision.The VoV-YOLO visual positioning method for rough castings takes into account the stability and measurement accuracy of the system,and improves the positioning accuracy of 2D vision system for rough castings.(2)When the 3D vision system is used to locate the rough machined metal casting,the local information is unstable in the imaging process due to the reflection of the workpiece,resulting in the deviation of pose measurement.A passive stereo vision bah visual positioning method is proposed,which effectively solves the adverse effect of incomplete information acquisition caused by the specular reflection of the rough machined metal casting on visual positioning.This paper constructs a multi key point detection network bah net,which can complete the 2D positioning of the left and right cameras of the stereo vision system at the same time,and provide reconstruction information for 3D pose measurement.Cycle GAN generation countermeasure network is introduced to enhance the local feature area image of the workpiece surface,and combined with RANSAC ellipse fitting algorithm and triangulation method to complete the three-dimensional pose measurement of the workpiece,which solves the problem of low accuracy of the vision system for the pose measurement of rough machined castings due to irregular reflection.(3)Multiple rough casts stacked disorderly are common scenes in industrial production.Due to the similar characteristics of the workpiece surface,the target recognition rate is low and the pose measurement error is large when the rough casts are positioned disorderly by the3 D vision system.Aiming at this problem,the PCA-PPF visual localization method of active stereo vision is studied.The binocular structured light algorithm is used to generate the three-dimensional point cloud of the measured workpiece,and the PPF point cloud registration algorithm is improved.The PCA-PPF registration algorithm based on the edge characteristics of the point cloud is proposed,which improves the target recognition rate of the active stereo vision blank casting positioning system,and realizes the grasping pose measurement of the industrial machine to the disorderly placed blank casts.The visual positioning method of industrial robot studied in this dissertation has high positioning accuracy and applicability.The proposed method has been applied in industrial field,which improves the production efficiency and automation level of enterprises,ensures the man-machine isolation in the production process,and achieves good economic benefits.