Research On UAV's Precise Landing On Ship Based On Vision Guidance

It has been an evitable trend for the development of Shipborne Unmanned Aerial Vehicle(UAV)due to its high performance but low-cost characteristics.However,compared with land-based UAV,more technical difficulties are involved in shipborne UAV,especially the automatic carrier landing guidance technology.This paper,anchored to the requirement of Shipborne UAV automatic landing technique,designed a vision-based ground guidance system,and focused researches of key technologies about target recognition algorithm in complex background,and construction of digital stabilization platform,which provided technical support for the engineering realization of shipborne UAV landing.Firstly,based on the landing features of shipborne UVA and the criterion of navigation system,a set of vision-based ground guidance system was designed followed by the analysis of its location evaluation principles and PTU tracking control,and a close scrutiny of the work procedure of the system.In addition,the flight-test platform of the shipborne UAV is designed,which paved ways for the flight test of the visual-based landing system.Secondly,UAV recognition algorithm based on HOG+SVM is devised,and the results are analyzed.To improve the recognition accuracy and enlarge the recognition range,a better robustness method named convolutional neural network was adopted to extract features,and UAV recognition algorithm was designed based on Faster-RCNN,which boosted the recognition accuracy up to 92.1%.Thirdly,to solve the problem of target scale change and real-time performance during the landing,a coalition algorithm was designed,in which frame difference was used for remote moving object detection,and Faster RCNN for short distance ones.In addition,focal loss algorithm was applied to the design to revise the limitation of loss function of Faster RCN,which solved the weighting problem of easy-classified examples and hard ones,and signaled to an improvement in recognition accuracy about 1.7%.As to the target recognition issues in image sequences,algorithm performance was refined according to continuity of motion,corresponded to the meeting of criterion for navigation system real-time performance.Finally,the flight test of the visual guidance system was carried out.For the beginning,the position calibration of PTU based on infrared vision is designed and tested.Furthermore,since deck movement makes PTU continuously move with the ship body,thus causing an unnecessary follow for the UAV to the deck.To solve this problem,a digital stabilized platform was constructed to eliminate the effects on PTU from carrier motion.The final flight verification results showed that the visual guidance system designed in this paper was highly feasible.