Research On Measurement Of Airborne Vision-based Navigation For Auto Landing UAV On Deck

Shipborne UAV can take off and land on the ship,and perform various tasks,which ensure it a broader range of operations than land-based UAVs.But involves more technical difficulties.The working band of visual navigation is far away from the current ECM frequency,which has the advantages of low cost,high precision,concealment and portability.Based on the research background of landing guidance recovery of UAV,we focus on vision guidance in the middle and late stages during landing on deck.Based on different visual features,different pose estimation algorithms are required by visual navigation.In the middle stage,clear and complete ship contour can be obtained from airborne vision,thus contour or edge based pose estimation methods are required;in the later stage,the whole ship contour can not be imaged,but the cooperative targets on the ship target can be detected.The technical difficulty of pose estimation based on point feature is how to stably arrange and detect cooperative signs on the ship in complex background.For the detection of visible photosynthetic markers in complex background,a timeencoded approach to detect cooperative targets for UAV landing is proposed.Visible light is used as the cooperative target of UAV landing,and a time relay is used to change the intensity of the light source.Difference image,floating threshold and geometric constraints are used to extract visible photosynthetic biomarkers from video sequence images stably.For the problem of point-based pose estimation with a few mismatches,by using the derivable property of Lie algebra and weighting the samples,an Lie-group-based iterative weighted pose estimation algorithm implemented.Experiments show that the proposed algorithm has the same accuracy as the classical RANSAC algorithm with a few mismatches,but has obvious advantages in computational efficiency.For the problem of contour-based pose estimation of monocular rigid objects,given the 3D model of the target,a rigid object pose estimation method in monocular images using 2D-3D matching is proposed.We analyzes the transformation equations of OpenGL virtual camera and OpenCV real-world camera parameters.Using the virtual camera as a bridge,we establishes the corresponding relationship from the input image to the target three-dimensional model indirectly.A framework of iterative pose estimation is designed,and the outter parameters of the virtual camera are updated using the method in chapter3.Experiments show that the proposed algorithm has high efficiency,strong robustness to noise and large convergence radius,and can be used for pose tracking in videos.