Precise Landing System For UAV Concentric Circular Targets Based On Machine Vision

With the development of image processing technology and the progress of embedded hardware,drone landing based on machine vision has become a very popular research field.There are high requirements for the stability,accuracy,reliability and real-time performance of UAV landing.The vision-based landing system is a hot spot for scholars.Divided from the traditional system,the system has the advantages of low cost and strong anti-interference ability.The basic requirement to realize the visual landing of drones is to obtain information about the environment in which the drone is located and use this information to accurately estimate the position and attitude of the drone.Among them,the recognition of landing targets,image processing and machine learning are the drones acquiring themselves.One of the methods of environmental information.This paper designs a hardware system based on machine vision for drones to accurately land,and studies target recognition and drone attitude estimation in the process of assisting drones in hardware and software systems.The main work content is as follows:(1)Designed a kind of redundant color sequence from the inside to black,red,black,green,black,blue,black,the length of the red ring is 10 cm,15 cm,and the size of the green ring is 20 cm.,30 cm,and the blue circle objects are 35 cm and 45 cm new landmarks respectively.This landmark not only satisfies the estimation of the drone's pose and is accurately recognized by the machine learning algorithm,but also ensures that the drone's airborne image sensor remote sensing image processing program can always detect the existence of a circle.(2)Researched the method of target recognition based on machine learning algorithm.In order to realize that the UAV can accurately identify the target with the aid of GPS coarse positioning.In this paper,four algorithms: U-net,random forest,multi-layer perceptron,and support vector machine are used to train the data set of concentric circles.The initial learning rate,batch size,and number of iterations are selected to target the target.Recognition,and finally the four algorithm models are analyzed on the obtained statistical accuracy table,and it is concluded that the random forest algorithm has the best recognition effect on the target.(3)Research on UAV pose estimation and attitude compensation method based on concentric circular target.This paper establishes a geometric model between the camera coordinate system and the image coordinate system by analyzing the imaging model of the airborne image sensor.Aiming at the problem that the traditional Hough transform in Open CV cannot recognize concentric circles,the method of converting between RGB color space and HSV color space and image processing on the concentric circle target is used to realize the recognition of concentric circles by the Hough transform function.Processing,extract the center of concentric circles.Use the relevant parameters of the image sensor and the altitude of the drone to obtain the attitude information of the drone,and finally control the drone to land by simulating the amount of the remote control stick.In addition,an attitude compensation method is used to compensate for the changes in the UAV's position caused by the inertia during the flight of the UAV.