The Research On Visual Guidance System Of Robot Based On Monocular Vision

Robot vision guidance is an important way for industrial robots to become intelligent.Traditional industrial robots can only do some monotonous and repetitive work in accordance with the written program,but can not adapt to adjust the trajectory of the robot according to the working environment.The visual guidance system gives the robot the ability to perceive the outside world.Based on monocular camera,this paper designs a 2D planar visual guidance system for the robot through the method of target recognition and feature point extraction.Finally,the system is tested by robot grasping experiment.The main research work of this paper is as follows:(1)According to the actual requirements of engineering system design,the overall scheme of robot visual guidance is designed and the hardware experimental platform is built.Aiming at the requirement of grasping and placing the workpiece accurately in the actual work process,the overall scheme of coordination between robot and flexible vibration plate controlled by vision system is proposed.The vision system obtains the posture information of grasping target through target recognition and attitude estimation,and guides the robot to grasp.After the robot completes all the target workpiece grasping,the flexible vibration plate is controlled to adjust the attitude of the remaining workpiece,and the cycle of photographing and detection is continued until all the workpiece grasping is completed.Finally,the hardware is selected according to the system scheme and the hardware experiment platform is built.(2)A target recognition method based on contour matching algorithm is proposed to complete the search of the robot grasping target on the original image,and to find the workpiece suitable for the robot grasping from the workpiece pile,excluding the lateral and stacked workpiece.In order to classify all grasping objects,this paper uses positive and negative template images for histogram comparison of each workpiece image to obtain two characteristic values,and finally completes the classification of the object workpiece by support vector machine.(3)In view of the low efficiency of SIFT feature extraction algorithm,this paper improved the algorithm to greatly improve the calculation efficiency of attitude.The pose estimation of the workpiece is mainly to solve the rotation Angle of irregular workpiece in the image,so it is necessary to extract the feature points of the workpiece.Improved SIFT algorithm: The gaussian scale space is calculated on the original image,without sampling,up and down through the adjacent two gaussian scale space image difference for gaussian difference image,then seek extreme value point in gaussian difference image as outline feature points,to outline the center of gravity as the center of rotation,with the template image,extraction of feature points are individually matching calculation of coincidence degree and sorting,The point with the highest degree of overlap is selected as the best matching point.Finally,the rotation Angle of the workpiece relative to the template image is calculated.(4)The feasibility of the visual guidance scheme proposed in this paper in the actual working environment is verified through the experiment of the robot grasping the workpiece,and the robustness of the image processing algorithm in the system in the actual environment is tested.Experiments show that the visual guidance scheme and image processing algorithm adopted in this paper can meet the application requirements of practical working environment.