Natural Feature-based Visual Servoing For Grasping Target With Aerial Manipulator

Due to compact,maneuverability and flexible mobility of the UAV,The UAV has been successfully applied in more and more fields.Recently,the applications of the UAV in aerial photography,power inspection and pesticide spraying have been quite mature.They mainly use the camera to provide a first perspective to assist in monitoring and detecting areas that people can’t reach.Based on existing algorithms and technologies,more and more researchers have empowered the UAV to actively interact with the external environment,such as aerial manipulator for grasping,cooperative transportation and human-computer interaction.The aerial manipulator system is a new research for the UAV.From the design and control of manipulator to the kinematics and dynamics modeling of complex system,the design of controller needs constant exploration by researchers.For the grasping problem,it is necessary to ensure stable,reliable and real-time performance.The traditional visual servo needs to be attached with artificial markers.However,in practical applications,the artificial markers can’t be attached due to the conditions.In order to solve these problems,natural feature based grasping is carried out.In this dissertation,the kinematics modeling of the six-degree-of-freedom manipulator is derived,and the forward and inverse kinematics model and the differential kinematics model are derived,which lays a mathematical foundation for the speed control of the aerial manipulator.Aiming at the assembly of the manipulator,the measurement error and the initial angle of the servoing motor,this dissertation proposes the calibration of the kinematic parameters of the manipulator based on the genetic algorithm.The mean square root error of the average position is 61.2% lower than that before calibration.The camera of the visual servo system adopts the "eye-in-hand" installation method and the image-based visual servo is the basic framework.In general,planar target can’t be attached with artificial markers.The dissertation proposes a visual servo control scheme based on natural features,the ORB natural feature is used for feature matching and affine transformation to obtain point features.At the same time,the problem of uneven distribution of features in the ORB detection algorithm is improved.Gaussian image pyramid is used to introduce the feature invariance of features.The quad-tree structure is used to store feature points.The distribution of the final feature points are uniform and sparse,which improve the visual servo effect.By modeling the differential kinematics of the aerial manipulator,this dissertation proposes a multi-task joint visual servo control strategy.When the UAV is far away from the target,the servo speed is transmitted to the UAV and one degree of freedom of the maipulator to make the pitch angle servo ensures that the camera can always see the target.When the manipulator reaches the workspace,only manipulator performs the visual servo task.In order to verify the effectiveness of the method,this dissertation builds an experimental platform based on the robot operating system(ROS),which can finish the autonomous grapsing of the aerial manipulator and doesn’t depend on the dynamic model.The proposed visual servoing method based on natural features can more meet the needs of real environment and it is impossible to stick markers on all the captured objects.