Development Of Dynamic Tracking And Pose Detection For Unloading Hole Of Reactor Base On Binocular Vision

With the rapid development of visual technology and the difficulty of information requirements in industrial processing,the visual dynamic measurement scene has changed from a simple static single image to a dynamic image under a complex background.Stereo vision measurement technology is an important basis for accurately extracting dynamic targets in complex background scenes.Combined with target detection and tracking technology,a reliable and effective visual dynamic detection and posture measurement system can be developed.At present,in the metal surface treatment process,large tank tooling is completed by manual operations.This kind of operating conditions has low working efficiency,harsh working environment,and serious automation level.It is urgent to solve and realize the tooling in the metal surface treatment production link.Automatic unloading.Based on the actual production process of the large tank clock,this paper effectively combines the current development status of stereoscopic inspection and online image recognition technology,compares the effects of each feature extraction model,and combines the convolutional neural network to image the dynamic large tank tool loading and unloading holes.The detection realizes the automatic recognition of the large tank clock;the surface treatment production line has a large environmental interference and high real-time requirements.Therefore,this article chooses to use YOLOv3 as the basis for research and improvement of the model.Reduce the amount of hyperparameter calculations,improve the speed of network detection,and compare it.This paper develops a stereo vision dynamic measurement system,compares the measurement accuracy of the four-dimensional key posture information of the space target before and after camera calibration,establishes an image Jacobian matrix conversion model for the base coordinate system and the image pixel coordinate system,and captures the image of the mechanical arm The Jacobian matrix uses the Kalman filter algorithm for online identification,the applicability of the Kalman filter tracking algorithm to the scene is studied,the real-time tracking controller of the grab arm based on the PID algorithm is designed,and the dynamic measurement system of the binocular vision is evaluated.Accuracy;using the gripping device at the end of the robot arm as a model,simulating the trajectory of the motion space for effective measurement and tracking,and the comparison results confirm the effectiveness of the system.Finally,the production conditions under specific industrial scenes were used as the operating environment,and the developed visual dynamic detection and posture measurement system were used to verify that the real-time detection of square holes under the interference of complex environmental coatings on the metal surface can be met.The rotating hole of the material tube,the angular positioning measurement and the posture positioning measurement can all achieve the precision of process automation,which confirms the availability of the system.