Research On Feature Extraction Technology Of Metal Highlights For Assembly Robot

The accurate acquisition of geometric features and parameters of mechanical parts is the premise and key of the assembly robot to complete the assembly task.Timely and accurate acquisition of the geometric features and parameters of the target parts can improve production efficiency and ensure product quality and production safety.Metal parts are widely used in actual production and assembly work,and their surfaces have special reflective properties to light.Due to the complex field environment and different light conditions,local or even large area of high light reflection is often produced when industrial cameras are used for image acquisition of metal parts,which makes the feature information of metal targets covered by the high light,resulting in the loss of target data and reducing the overall quality of the image.In this paper,the following research is carried out on the situation that highlights appear in the image of metal parts in the highlight environment in the assembly operation of the assembly robot,which covers the geometric features of the target and leads to low detection rate:1)In order to narrow the search scope of target extraction and reduce the influence of complex background factors,the YOLOv5 target detection algorithm is proposed to be applied to the rough location of parts targets.The experimental results show that the m AP value of the proposed model is up to 96.6% for gasket recognition and 95.9% for gear detection,which proves the accuracy of the proposed model and meets the needs of actual target detection.2)Aiming at the problem that highlights appear in the image of metal parts during image acquisition,covering the geometric features of the target and resulting in low detection rate,a highlight extraction and recovery algorithm based on the combination of saliency detection and fast moving repair is proposed.The restoration effect is evaluated by the PSNR value and SSIM value.After processing by the algorithm in this paper,the PSNR value of the image is increased by 9.8%,while the SSIM value is reduced by 28.1%,effectively restoring the highlight information in the image.3)In view of the problems of weak robustness and high requirements for ambient light in the extraction and positioning of target features to varying degrees,the least square fitting ellipse method is adopted in this paper to replace the least square fitting circle.Finally,it is compared with gray gravity center method and Hough transform method.The results show that after the removal of highlights,The first two algorithms can keep the positioning accuracy within 2.5 pixel,while the proposed algorithm can keep the center deviation within 0.3 pixel,which has high accuracy and stability.4)Construct the feature extraction system of metal parts in the highlight environment.A feature extraction system for metal parts highlight images of assembly robots was developed,which has the functions of user login,rough location of target parts,highlight recovery of target metal parts,feature extraction of target and display of original and result images,etc.The effectiveness of the proposed algorithm was verified.