Control System Of Robot Arm Based On Deep Learning With Binocular Vision

With the improvement of automation level and production scale,modern industrial production line is developing to the direction of flexibility and intelligence.As the actuator on the production line,the inherent requirements of the autonomous environment recognition ability and grasping precision of the robot arm are also increasing day by day.However,the vast majority of industrial robot arms still show that the way of teaching works according to the predetermined trajectory,when the work environment or task changes,it is unable to make timely adjustments,and the intelligent level of its control system is difficult to meet the development needs of modern production lines,which is an important problem in the current industrial field to be solved urgently.In recent years,with the development of hardware and the continuous proposal of new algorithms,deep learning technology and machine vision technology have made a breakthrough,and have been widely used in aerospace,face recognition,real-time visual translation,automatic driving and other fields.Will,therefore,in-depth study and combined with machine vision technology and the mechanical arm control system,give full play to its own identification and the target object feature points the advantages of accurate extraction has important research significance and broad application prospect,for improving the mechanical arm of the independent environment awareness and grasping motion precision,this article launches the research from the following three aspects:(1)The target recognition algorithm based on deep learning.First is the design and implementation of deep learning network,and then set up a data set and the target object using built deep learning network training of target recognition model,finally will contain the target of binocular camera taken photos of the object and jamming to deep learning network and identification of target object,extract significant area,within the scope of the test box into subsequent algorithm for image processing.(2)Based on image processing,the two-dimensional coordinate extraction algorithm of the grasping point of the target object is proposed.First to extract significant area of the image processing,including image denoising,threshold segmentation,feature point extraction,etc.,and then use have to extract the feature points on the main part of the target object image centroid settlement,using the feature points of the centroid coordinates of the maximum inscribed rectangle as the grasping point coordinate,to ensure that grasping some basic located above the center of gravity of the target object,to ensure that the mechanical arm can effectively complete fetching.(3)The Three-dimensional reconstruction method based on binocular stereo vision theory.First use a combination of inside and outside ilf camera calibration method of camera parameters,with parallel optical axis of the theory of binocular stereo vision range finding principle and the similar triangle principle of parallel optical axis model,then grab some two-dimensional coordinates of target body stereo matching and Threedimensional reconstruction,calculate the point coordinates in the world of Threedimensional coordinate,the final conclusion by coordinate transformation that point in the mechanical system of three-dimensional coordinates,and incoming manipulator path planning and grasping motion.In order to verify the feasibility and effectiveness of the above studies,an experimental platform of the manipulator control system was built,including the selection of hardware equipment and the writing of program code,and the experiment of object recognition and grasping,recording and analyzing experimental data.The experimental results show that the average target recognition time of the manipulator control system is only 0.87 seconds,and the total grasping time is only 7.85 seconds,which verifies the accurate and fast grasping ability of the manipulator control system and its robustness in different working environments.