Research On Robot Grasping And Locating Control Technology Based On Stereo Vision

Robots were widely used for their advantages such as high precision,high efficiency,resistance to harsh environments,and continuous work,in the field of industrial manufacturing.With the continuous improvement of industrial automation,traditional robots can no longer meet the more complex automation requirements at this stage.Integrating the vision system on the robot has become a new mainstream direction for robot development.The currently common monocular vision robot can only posit and grab on a two-dimensional plane,and binocular stereo vision can calculate the three-dimensional information of the surface of the object in the field of view according to the acquired image,which is complicated The scene provides high-precision pose information to support the robot's subsequent grasping work.Based on the existing problems,this paper analyzed the theoretical basis of target recognition and 3D locating,and studied the point-by-point teaching process of traditional robots' "grab-place" actions based on visual systems,including: recognition of target objects and grab points of target objects.The placement point of the target object and the trajectory between the grabbing position and the placement position,optimized the existing related algorithms including calibration algorithm,recognition algorithm,etc.Finally,based on QT,Visual C ++ and OpenCV as development tools,designed experiments to verify.Including scenarios the main contents of data processing processes such as construction,object recognition,and trajectory calculation are as follows:(1)In order to improve the recognition accuracy of the robot,a camera calibration method based on stereo camera driven by the pan-tilt is proposed.In view of the imperfection of the existing monocular camera in obtaining scene information,rotating gimbal can used to drive the camera to complete any rotation of the camera to shoot the scene.The gimbal rotation process will produce a pitch angle,which is then based on the angle of the gimbal rotation,eliminated the error to realize the external parameter estimation of the camera and completed the entire calibration process of the camera.Then,for the above calibration method,an efficient target segmentation and recognition method was proposed to improve the accuracy and efficiency of the target object recognition.The experimental results showed that the camera calibration method optimized by external parameter estimation in this paper can greatly reduce the data error and maintain the rotation efficiency of the gimbal.(2)An improved feature extraction strategy was proposed.Aiming at the problem of incomplete features obtained by traditional segmentation algorithms,on the basis of traditional segmentation algorithms,a deep learning strategy was added,and feature extraction and segmentation were completed by using convolution methods.First,using the RPN network to select the Region as the prefetch area of the data.Then,using the Mask prediction branch to add improved ROI Pooling for identification and classification.Experimental results showed that the network model proposed in this paper can meet the needs of feature prefetching to a large extent,and was higher in time efficiency and accuracy of recognition than traditional methods.(3)Designed and developed a carton recognition and locating detection system based on machine vision.Using different sizes of carton size data,the optimization methods proposed in this paper were compared and verified,and the carton locating coordinate solution function was implemented in the system.The experimental results showed that the position information obtained by this technology can quickly capture the target in the region of interest through testing and motion trajectory planning,which can significantly improve the accuracy and timeliness of object recognition under scene changes.It has a good assisting effect on industrial automation production lines,which is conducive to saving human resource costs and enhancing the intelligent development of industrial manufacturing.