Research On Visual SLAM Algorithm Of Picking Robot Based On Depth Camera

Picking robots have a very important role in realizing agricultural automation,and the new smart agricultural production model has promoted research on related technologies.In recent years,the development of picking robots mainly includes autonomous mobile technology,manipulator design,and detection and identification of picking objects.Among them,autonomous movement is the most basic and most important capability of picking robots during the production and harvesting process.Simultaneous localization and mapping(SLAM)combined with machine vision has become a research hotspot.In order to improve the accuracy and robustness of picking robot positioning and mapping,this paper studies the picking robot’s pose estimation and pose optimization algorithm,and designs a visual SLAM system based on depth camera.The main contents are as follows:Learn the theoretical framework and basic algorithms of visual SLAM technology,and develop a visual SLAM system that uses a depth camera as a perception device.Master the principle of Kinect v1 to obtain color pictures and depth pictures.The model establishment,visualization and motion control are completed based on the Robot Operating System(ROS).In order to solve the pose estimation problem of the picking robot at the front end of the visual SLAM,a quadtree-based ORB image feature algorithm and pose calculation method are proposed.Based on the ORB image detection algorithm,for the problem of uneven feature point extraction in the detection,the quadtree uniform processing is used to achieve the purpose that the extracted feature points can be evenly distributed in the image.After this,the fast matching of feature point pairs and the elimination of false matching can be effectively promoted.Through good data matching information,fusion of the idea of random sample consensus algorithm,which obtain the optimal pose after multiple iterations by combining EPNP and iterative closest point method.Experiments with data from different scenarios verify the beneficial effects of the improved algorithm.Aiming at the problem of inaccurate pose estimation caused by cumulative errors,a visual SLAM-based picking robot pose optimization algorithm is proposed,which can achieve the consistency of the global trajectory.Introduce the basic theory of graph optimization,calculate the close-loop information based on the similarity calculation using the bag of words model.Moreover,we design the selection criteria for keyframes to optimize the data size and select the keyframe set.With the pose variables to be optimized as nodes,the estimated pose constraints and close-loop constraints between nodes as edges,the g2 o library is used to construct and optimize the pose graph structure.The experimental results show that the performance indicators of the optimization algorithm are significantly improved,and the trajectory estimation error is reduced.Design the overall framework and algorithm flow of the visual SLAM system to realize the research of the picking robot in the simulation and experiment platform.In the case of simulation,the Gazebo plugin in ROS is used to design the physical simulation environment,and the movement of the picking robot model is completed through the motion control mode of the publishing node to verify the reliability of the visual SLAM system.The experiment of positioning and mapping of the picking robot in the real environment was carried out,and compared with the RGB-D SLAMv2 method,the results showed good positioning and mapping effects,indicating that the system in this paper is more robust and accurate.