Research And Experimental Validation On Target Tracking Based On Under-actuated Robotic Arm End Vision System

With the increasing demands for artificial intelligence and smart manufacturing,the Intellectualization of control systems is becoming a research hotspot.To improve system intelligence,in this work,an under-actuated robotic arm platform featuring intelligent target detection and tracking is designed and verified.Firstly,a visual perception module for image feature extraction is proposed.Based on the analysis of camera imaging characteristics in the experimental environment,target detection algorithms based on both traditional filtering and convolutional neural networks are designed.The former costs less time,but is not effective in the detection of multiple targets with color close to the background,while latter requires specific software and hardware support.After the process of image pre-processing,target detection,feature point extraction and update,this module can realize the effective extraction of target feature points in complex environment,and the algorithm can reach30 FPS,which meets the requirements of both accuracy and responsiveness.Then,a position-based visual servo scheme is designed.It is able to settle the target spatial position according to the change of target feature points,and then derive the control expectation.However,this scheme can not meet the real-time requirements.Due to the loss of monocular vision depth information,significant error can be found in the derived spatial position of the target,and the time complexity of the system is unaffordable.To overcome the shortcomings of the above position-based vision servo scheme,an image-based vision servo scheme is proposed.The scheme directly defines the control error in the image space without solving the target spatial position,which can speed up the algorithm.Additionally,solution to moving targets is optimized in this scheme,leading to the faster convergence of the algorithm.Besides,error compensation for system delay is done with the Kalman filter,while error of the position of the feature points caused by motion coupling of the under-actuated system is also compensated.The algorithm can converge rapidly to meet the real-time requirements and is of good robustness.Finally,an under-actuated robotic arm platform is established and the target tracking algorithm is experimentally verified.An under-actuated robotic arm physical platform with a monocular camera,integrated with the aforementioned vision perception module and vision servo module,is built and tested.The experimental platform is also compensated with a uncertain and disturbance estimator.Then,simulation experiments on stationary and moving target tracking are designed in MATLAB in combination with the control model of the platform,and physical experiments on mobile robot tracking are arranged.Both the simulation and the physical platform test results show that the detection and tracking algorithm designed in this paper for moving targets is effective,and the under-actuated robotic arm platform can be combined with monocular vision to realize the intelligence of the platform.With the proposed under-actuated robotic arm platform combined with machine vision technology,the autonomous tracking of targets of the robotic-arm-end can be well realized.Furthermore,the platform act well both as simulation and experimental examples of vision control theory,which can be valuable in further application and research.