Research On Workpiece Recognition And Tracking Method Based On Monocular Vision

With the continuous improvement of the production requirements of the factory products,the rational use of robotic equipment can not only speed up the production speed of the products and improve the production efficiency,but also help people to complete some production tasks that are difficult to complete by humans and have complicated production processes.In the traditional sorting process of the enterprise,manual sorting or mechanical sorting is generally used for the case where the number of products is large and the types are large.Aiming at the complicated sorting problem of production line products,a monocular vision workpiece recognition and tracking system was designed,which can replace the manual sorting and mechanical sorting operations in industrial automation production to improve production quality and efficiency.The main work done for this system is as follows:In order to ensure the stability and efficiency of the monocular visual recognition tracking system,in the selection of the robot type,the SCARA robot is used in the experiment as the actuator for sorting the workpiece,and the monocular camera is used as the image acquisition of the workpiece.The equipment use camera and SCARA robot with the conveyor belt as the transmission mechanism to build the entire identification tracking platform based on monocular vision technology.First,for camera lens distortion problems with monocular cameras,the camera needs to be calibrated.After comparing the advantages and disadvantages of different calibration methods,the experiment uses Zhang Zhengyou calibration method combined with the calibration assistant and the packaged operator of the Halcon platform to calibrate the camera,thus obtaining the internal and external parameters of the camera.The calibration algorithm is simple and easy to operate.The calibration result is highly accurate.Secondly,in order to reduce the influence of dim light and noise pollution on the industrial site,and to ensure the image quality,the experiment selects the appropriate filter and edge detection operator to denoise and edge detect the workpiece image respectively,which reduces the later The complexity of the workpiece identification tracking.Then,in order to make up for the shortcomings of the traditional target detection and recognition algorithm to the external environment change,such as weak adaptability to external environment changes,poor anti-disturbance ability,and dependence on the light source,a method of combining template matching with neural network is proposed to realize the detection and recognition of the workpiece.Firstly,the actual captured workpiece image is pre-matched with the template library contour to find the workpiece that fits the image contour matching parameter range.Then,the MLP neural network is used to train the contour of the workpiece contained in the template library,and the pre-matching does not meet the requirements.The workpiece is subjected to secondary matching to achieve multi-objective recognition classification and improve the recognition rate of the workpiece.In the aspect of moving workpiece tracking,a background difference method and Kalman filter tracking algorithm are proposed to predict the range of possible motion artifacts,reduce the search range when correlation is matched,improve tracking accuracy,and focus on the coordinate conversion formula.The coordinates are converted to grab point coordinates.Finally,based on the Halcon platform to complete the identification tracking system framework,the workpiece recognition tracking human-computer interaction interface is designed on the VS2017 platform.