Research On Target Recognition And Grabbing Of Double Manipulator Rotorcraft

Rotorcraft has the advantages of small size,strong maneuverability,and simple operation.Therefore,it is widely used in aerial photography,reconnaissance,and mission-oriented operations.In particular,a rotorcraft with a robotic arm can detect and identify targets,perform visual positioning,and manipulate the robotic arm to grab objects when performing grabbing tasks.Therefore,the rotorcraft with a robotic arm based on target recognition and visual positioning technology has become a current hot research direction.This article studies this research direction and designs a rotary drone with a dual robotic arm.The main work includes design experimental platform,SSD-Mobile Net algorithm research,visual positioning method research,dual manipulator motion method research:(1)In response to the mission requirements and working environment of this paper,an experimental platform for a dual-robot rotorcraft was designed.Analyze the performance indicators of the main hardware modules,use Intel NUC as the onboard computer to run the corresponding programs,use Pixhawk as the flight control,use ROS system to write the module programs and package them into software packages;finally give the system work structure block diagram and overall Software work flow chart.(2)When performing a grab mission based on a rotorcraft,the target object must first be identified.It is found that the traditional method of target detection and recognition based on HOG features is not applicable in the context of this article.Instead,research on the method of target detection and recognition based on convolutional neural networks.From some common target recognition algorithms,it is determined that the SSD network is used as Based on this,Mobile Net algorithm is added to reduce the amount of calculation and model size,and an SSD-Mobile Net fusion algorithm based on SSD and Mobile Net is proposed.Finally,experiments show that the method can effectively recognize the objects in the scene when it is applied to the rotor drone.(3)Investigate two kinds of visual positioning methods: visual positioning based on depth camera Real Sense R200 and visual positioning based on April Tag.Firstly,the basic principles of the two methods are studied.The former obtains the 3D coordinates of the target with respect to the camera through the IR image and the depth image,and the latter calculates the 6D information of the label through the homography matrix and the camera internal parameters.Several sets of comparative experiments were done on the experimental platform,comparison shows that the latter has higher positioning accuracy,Therefore,it is determined to use the April Tag-based visual positioning method to obtain the three-dimensional coordinates of the target,and the visual positioning of the target is achieved through experiments.(4)The basic principles of the RRT_Connect algorithm and the FCL collision detection library are studied,and then the three-dimensional coordinates of the target relative to the camera are converted to the coordinates relative to the double robotic arm.The ROS system is used to configure the dual manipulator so that it can be controlled by the program.The inverse kinematics of the dual manipulator is solved,and the rotation angle of each joint of the dual manipulator is planned to move to the target position in three-dimensional space without collision.