Vision Navigation Technology Based On Multi-source Information Fusion For UAV

With the development of high-tech such as artificial intelligence and automatic driving,unmanned aerial vehicles(UAV)have become the hotspot of current research.The UAV can land on its own or through the block,and can complete the tasks such as take-off and navigation autonomously.The landing phase is the high-risk phase.Therefore,this paper proposes a visual aided UAV autonomous landing system based on multi-source data fusion under GNSS-denied conditions.The main research contents of this paper are as follows:(1)Firstly,the new landing targets are designed from the aspects of shape,size,color and material.Then,the performance difference of the T-shaped,the H-shaped typical landing landmarks and the landing landmarks designed in this paper are compared through experiments.(2)For the landing landmarks designed in this paper,different methods of visual processing are proposed.For UAV long-distance landing,this paper proposes an image preprocessing method based on RGB color space.For the UAV close-range landing,this paper proposes an image preprocessing method based on HSV color space.Then the edge detection algorithm based on Canny algorithm is used to identify the landing landmarks,and the performance of different feature point detection algorithms is compared.(3)In this paper,the feature point tracking method based on pyramid LK optical flow method is proposed to ensure the accuracy and robustness of the UAV's autonomous landing tracking system,and to achieve stable and fast tracking of the ship's landmarks.(4)This paper introduces the common coordinate systems involved in the visual pose estimation of the drone and their conversion relationships.The camera calibration provides the camera internal parameters for the visual pose estimation.Finally,the Pn P algorithm is combined with the internal parameters of the camera to calculate the posture of the drone.(5)In this paper,the cumulative error characteristics of inertial navigation system are introduced,which provides a theoretical basis for the simulation of INS error data.TheKalman filter model is designed to realize the local positioning technology based on the multi-source information fusion under GNSS-denied conditions.The INS cumulative error intermittent correction scheme is designed to improve the positioning accuracy of the UAV visual navigation system.This paper designs a three-dimensional visual simulation and verification system.The three-dimensional visual simulation system combines with the algorithm simulation,and simultaneously displays the three-dimensional visual field and real-time visual navigation pose estimation parameters of the UAV,and verifies the feasibility and effectiveness of the proposed algorithm.Finally,the results of the demonstration verification system are analyzed.The system can realize multi-channel and multi-field dynamic three-dimensional linkage visual display processing,and realize the unification of three-dimensional visual presentation and algorithm simulation verification.Finally,the real-time performance and accuracy performance of the system are analyzed.The UAV autonomous landing system of this paper can realize the accurate landing of the UAV and meet the real-time requirements.