Research On UAV Visual Obstacle Avoidance Technology Under GNSS Constrained Conditions

With the rapid development of drone technology,the flight environment and missions it faces are becoming more and more complex,which puts forward higher requirements for the autonomous flight of drones.However,the loss and weakening of GPS signals will seriously affect the safe obstacle avoidance flight of drones.In response to this situation,drones can only rely on various onboard sensors to complete the perception of the surrounding environment and estimate their own pose,and then complete tasks such as autonomous obstacle avoidance flight.Technology is the key technology to achieve obstacle avoidance flight.Among the various types of sensors,the cost of the camera is low,the data collected is rich in information,and has wide adaptability.This paper aims at the obstacle avoidance flight phenomenon of drones under the condition of limited GNSS,and studies the fusion positioning technology based on monocular vision and inertial sensors and the obstacle avoidance technology based on this,to realize the vision-based drone The fusion positioning with inertia has accumulated the necessary data for its obstacle avoidance flight.The main work of the paper is as follows:1.The image frame processing technology and the posture calculation technology based on monocular vision are studied.Explains the camera projection model,introduces the conversion relationship between each coordinate system and the key technology commonly used in visual positioning technology;then analyzes the classic feature point extraction algorithm,and implements an improved ORB-based feature point based on the research content of this article The extraction algorithm achieves the uniform extraction of feature points.In terms of vision-based pose calculation,the mathematical model of this task is studied,and two pose estimation schemes based on monocular vision are proposed.One is based on inter-frame motion estimation,and the other is based on Based on the motion between frames,map points are added to participate in the scheme of optimizing the pose.Through the simulation environment and public data set,the performance analysis and comparison of the two are completed,which provides the basis for the design of the visual odometer.2.This thesis studies the positioning system based on vision and other sensors,and designs the UAV obstacle avoidance scheme based on this positioning technology.Analyzed many important links of positioning system construction,such as visual initialization,partial mapping,etc.,realized the use of key frame technology to improve the operating efficiency of the system.Since monocular vision will cause the loss of scale in the positioning system,an optimized vision and inertial fusion solution is studied to solve the problem of scale loss.The sensor data fusion process uses IMU pre-integration and other technologies to realize the scale of monocular vision.restore.In addition,a vision-based obstacle avoidance scheme is designed,and a technology that combines fixed-focus ranging method and visual-inertial fusion positioning method for UAV visual obstacle avoidance is proposed,which provides a theoretical basis for the following experimental verification.3.This thesis studies the design of simulation experiment scheme based on ROS robot operating system,introduces the contents of experiment scheme,scene construction,data communication and camera calibration,etc.,and also realizes the design of visual interface completed by Open GL.The scene is built using the three-dimensional simulation platform Gazebo,which is highly compatible with ROS.Finally,a simulation environment was built to simulate the vision and inertial positioning system and obstacle avoidance technology proposed in this paper,and the effectiveness and accuracy of the vision-based positioning system and obstacle avoidance scheme were verified,and the vision and inertia based Posture estimation and obstacle avoidance flight...