Research On Robot Arm Positioning And Tracking Algorithm Based On Visual Guidance

With the rapid development of technology,robotic arms have been widely used in industrial manufacturing,medical,military and other fields.However,the positioning and tracking issues of robotic arms have always been one of the bottlenecks that constrain their application.In order to improve the accuracy and stability of robotic arms,vision guided robotic arm positioning and tracking algorithms have become a research hotspot.This thesis focuses on the research of robot arm positioning and tracking methods based on visual guidance in unstructured environments.The main content is as follows:Firstly,the visual system of the robotic arm was analyzed,and an image denoising algorithm based on improved two-dimensional empirical mode decomposition and Wiener filtering was proposed to address the problem of high image noise during the positioning and tracking process of the robotic arm.By using the improved two-dimensional empirical mode decomposition technology,the noisy image can be effectively decomposed into noisy intrinsic mode function components,and combined with the Wiener filtering technology,the maximum noise reduction can be achieved.The experimental results show that this algorithm can effectively remove image noise and improve image quality,and the denoising ability of the algorithm proposed in this thesis is superior to other comparative algorithms for high noise images.Secondly,in response to the complexity and low robustness of traditional image processing and feature extraction processes,this thesis adopts a deep learning based object tracking detection method and proposes an improved YOLOv5 s and Deep SORT based object tracking detection algorithm.The improved YOLOv5 s model is used as a detector for Deep SORT to improve tracking performance,reduce target identity switching rate,and improve the accuracy of target feature information by improving target detection performance.Then,the method proposed in this article was tested,and the results showed that compared to the original YOLOv5s+Deep SORT,the improved YOLOv5s+Deep SORT model improved MOTA by 13.1%,MOTP by 3.3%,and had good tracking and detection performance for occluded and small-scale targets,providing effective target feature information for the planning and control of subsequent robotic arms.Finally,a maximum correlation entropy Kalman filter filter algorithm based on RBF neural network is proposed to estimate the image Jacobian matrix in the uncalibrated robot arm target location method.In this thesis,the maximum correlation entropy criterion is introduced into the Kalman filter framework to suppress the influence of non Gaussian noise on the filtering accuracy in robot visual servo,and then RBF neural network is used to compensate the estimation error generated by the maximum correlation entropy Kalman filter filter algorithm.The simulation and experiment results show that the accuracy and robustness of visual servo positioning can be effectively improved by improving the accuracy and stability of Jacobian matrix and determinant estimation.In simulation experiments in non Gaussian environments,the proposed algorithm can control the positioning error within 1.9 pixels,and the image feature trajectory is relatively stable,with strong anti noise interference ability.