Research On Vision Assisted Autonomous Navigation Of UAV In Wide Area Environment

With the development of UAVs’ application in various industries,the requirements for UAVs’ autonomy are increasing.Particularly,performing complex tasks in unknown environment is an important aspect of the research.Autonomous navigation is an effective guarantee to improve the autonomy of UAV.The application of computer vision technique and vision sensors improves the level of autonomy and intelligence of UAV in flight.Especially in the environment of satellite navigation signal blocking,visual sensors have been widely used in flight because of the strong autonomy.On the one hand,the attitude calculated based on the information output from the visual sensors does not contain drift,which can make up for the drift of inertial navigation information and improve the performance of autonomous navigation for UAV,but the information fusion of the two sensors also faces many challenges;On the other hand,the environmental information collected by the visual sensor can also assist the UAV to determine the flight route in the flight environment,select the landing location,adjust the posture during landing,and land in the designated area accurately.This paper takes the rotor UAV as the research object,and carries out the research on vision assisted autonomous navigation and flight of UAVs in wide area environment.Firstly,aiming at the wide and unknown environment,the research on vision assisted autonomous navigation of UAVs independent of satellite navigation signal is carried out.Aiming at the problem of vision failure caused by sparse texture and missing features in some areas,a visual-inertial navigation method is proposed,which combines the pose information calculated by minimizing the gray difference of different images with the inertial navigation pre-integration results.It improves the utilization of image information collected in the wide environment with sparse features,and enhances the reliability of visual navigation.The test results on public data sets show that the fusion navigation algorithm proposed has good navigation accuracy in wide environment.Secondly,aiming at the problem of quickly approaching important targets along the unpredictable roads in wide environment,aiming at the problem that there is no unified method to determine local flight path points when UAV follows the road with different shapes and corners,a tracking approach navigation method based on minimizing the angle between velocity and target vector is proposed.The proposed algorithm is verified in the simulation environment.The results show that the coincidence error between the projection of UAV tracking flight trajectory and the actual road is less than 0.1 meters,and there is no need to significantly change the UAV attitude,which greatly improves the efficiency and accuracy of tracking flight.Finally,aiming at the demand of autonomous landing of UAV in the automatic hangar rapidly and accurately after returning from the wide environment,an autonomous navigation method for UAV landing assisted by a small number of images including integrated signs is proposed.In an ideal situation,based on the results of identifying the mark and information processed on two images,the attitude and position of UAV are adjusted successively for autonomous landing.The UAV platform is selected to verify the practicability and rapidity of the algorithm.The average radius of landing error is 0.15 meters,and the yaw angle error is less than 10 degrees.