Research On Autonomous Obstacle Avoidance Technology For UAV Based On Monocular Vision

With the rapid development of related technologies,UAV has been widely used in various industries in recent years.However,the traditional way of controlling the movement of UAV by human limits the process of UAV autonomy and intelligence and its further wide application.In the intelligent applications of UAV,it’s a significant and challenging subject for autonomous systems to realize safe and reliable obstacle detection and collision avoidance of UAV.The paper aims at problems of autonomous obstacle detection and collision avoidance of UAV in complex scenes.Using the monocular camera carried by UAV,this paper studies the autonomous obstacle avoidance method based on monocular vision.The autonomous obstacle avoidance system is divided into two parts: autonomous environment perception system and UAV autonomous motion controller.The autonomous environment perception system is responsible for sensing the environmental information around the UAV and detecting obstacles in the moving direction.The UAV autonomous motion controller controls the UAV motion according to the obstacle information detected by the autonomous environment perception system.The main research contents of this subject are as follows:Firstly,the framework of UAV autonomous obstacle avoidance system based on monocular vision is designed.Aiming at the small quadrotor UAV,the autonomous obstacle avoidance method of UAV based on monocular vision was studied in order to meet the real-time and accuracy requirements of UAV obstacle detection under the condition of medium and low speed.The imaging principle of monocular camera is explored to lay a theoretical foundation for the subsequent distance estimation based on monocular camera.The basic theory of obstacle detection method based on monocular vision is introduced to provide theoretical support for the construction of autonomous environment perception system.Due to the large size of the original image collected by UAV,the real-time performance of the autonomous environment perception system will be reduced.In addition,the edge part of the image has a low impact on the flight of UAV.Faced with the above problems,a pre-processing method of the original image is explored to cut out the edge part of the image.Then,the autonomous environment perception method based on monocular vision is studied.In order to meet the real-time requirement of UAV’s environmental perception as much as possible,the performance of three common feature point extraction algorithms is tested and compared.The obstacle state determination method based on visual expansion is studied.On this basis,the multi obstacle detection method based on monocular vision is studied.The multi obstacle distribution area is detected by introducing ISODATA dynamic clustering and YOLO-v3 target detection.Next,for the safe and stable UAV control,the autonomous motion control method of UAV based on the distribution of multiple obstacles is studied.In order to quantitatively describe the distance between obstacle and UAV,a distance calculation model based on monocular vision is established to estimate the distance.A PID controller is designed to perform autonomous obstacle avoidance of UAVs in a scenario with known multiple obstacle distributions.In order to make the UAV fly more smoothly,a fuzzy controller is designed to integrate the distance information between the UAV and the obstacle and the real-time speed information of the UAV to control the steering angle of the UAV.Finally,a simulation experiment is designed to verify and analyze the real-time performance and accuracy of the autonomous environment perception system.And the autonomous obstacle avoidance ability of the system is tested in the simulation environment.Experiments demonstrate the effectiveness of the system.