| With the development and progress of industrial production technology,there is an increasing demand for intelligent robot-related technologies.As a core function of intelligent robots,machine vision plays a crucial role in industrial production.However,traditional machine vision technology in industrial production processes is no longer able to meet the demands of flexible production line-related tasks.This paper mainly investigates the use of image data collected by the Kinect V2 depth camera to achieve object detection and 6-Do F pose estimation,and combines it with robotics to achieve object grasping.The work in this paper includes several aspects:Firstly,the imaging principle of the depth camera was investigated,and the camera was calibrated and the distortion caused during the imaging process was corrected based on mathematical relationships.The calibration principles of the camera and the potential distortions that may occur during its usage were explored.The Zhang’s calibration method was employed to calibrate the color camera and depth camera of the Kinect V2 camera,resulting in intrinsic matrices for both cameras.The collected depth and color images were aligned using the Open CV library functions,yielding a transformation matrix between the color and depth images.Secondly,target detection was achieved through a template matching algorithm,and it was improved in practical applications.Using Open GL to capture target model images from different perspectives and record their pose information,the template image features were extracted and saved in a template library.To address the issues of occlusion and repeat detection in Linemod algorithm,a local sampling-based template matching algorithm was proposed,and a new similarity evaluation function was introduced.Finally,the target detection experiment was completed,validating the effectiveness of the algorithm.Thirdly,the pose of the target workpiece was estimated.Firstly,the collected depth images were preprocessed.The pass-through filter was used to remove the points irrelevant to the target object,and the outlier points far away from the target were removed through statistical methods.The process of pose estimation includes two stages: rough registration and fine registration.In the rough registration stage,the pose information recorded in the template library was used as the initial condition for accurate registration.In the fine registration stage,the generalized iterative closest point algorithm(GICP)was utilized to obtain the 6-Do F precise pose of the target object.Finally,A robot grasping experimental platform was designed,and the effectiveness of the proposed algorithm was verified through simulation experiments.The practicality of the developed vision system was also demonstrated in real experiments.The software modules included an image acquisition module,an object recognition module,an object pose estimation module,and a robotic arm grasping planning module.The hardware components consisted of a robot,a depth camera,and an end effector.The hand-eye calibration of the system was performed,and robot trajectory planning was conducted using Move It! software.Grasping experiments on target objects were carried out in both the Gazebo simulation environment and the real environment,yielding favorable results.The experimental outcomes indicate that the proposed method can successfully accomplish robot grasping experiments. |
