Research On Indoor Positioning Algorithm Of Quadrotor Aircraft Based On Vision

Quadrotor UAVs are widely used in surveying and mapping,aerial photography,disaster monitoring,search and rescue,etc.Using UAVs to carry out complex and dangerous tasks can improve work efficiency and reduce casualties.Due to its huge market and application prospect,the autonomous navigation of the quadrotor UAV in the unknown environment has become a hot research direction of all countries.Classical visual SLAM has many difficulties and challenges in the application of unmanned aerial vehicles.The most representative methods include feature point method and direct method.Among them,the classical feature point method has a relatively mature technical scheme,but it requires high image quality.When the image is low texture or fuzzy,the visual tracking is easy to fail.In addition,the feature extraction and the calculation of descriptor will consume a lot of time.The direct method uses the photometric information of the image to estimate the pose of the image,without the need of feature extraction and descriptor calculation,but the direct method has strict requirements on illumination conditions.Therefore,by combining the advantages of the feature point method and the direct method,this paper proposes an improved binocular vision SLAM system based on the semi-direct method,which enables the algorithm in this paper to achieve the effect of real-time processing and estimate the spatial depth information of three-dimensional points through binocular vision,which better improves the shortcomings of the above methods.The main work contents of this paper are as follows:(1)The mathematical description of the SLAM system is carried out,the commonly used coordinate system definitions are analyzed,and the transformation relations among the camera coordinate system,image coordinate system and world coordinate system are given,and the detailed mathematical derivation formula is given.Based on the matching relationship of image feature points,the essential matrix can be solved,and then the rotation matrix R and the translation vector T can be decomposed to derive the motion of the camera.In addition,the challenges and difficulties encountered in the application of visual SLAM in UAV are presented.(2)Aiming at the shortcomings of the feature method and direct method,an improved binocular SLAM system based on the semi-direct method is proposed.First,the image alignment of the binocular camera is carried out,and then the initial pose of the camera is obtained by using the direct method to optimize the photometric error.Then,using the advantages of ORB feature points,the left and right images of the camera were extracted,and the depth information of the feature points was calculated through binocular vision information.Then,combined with the image pose obtained in the previous step,the reprojection error was optimized to reduce the error matching and improve the robustness of the system.(3)In order to verify the validity of the algorithm,this paper gives the hardware and software of the system framework,in the simulation environment,Eu Ro C data set and the actual uav platforms were tested,through assessment and analysis of the experiment,and compared with mainstream open source SLAM system and verified in this paper,the algorithm has good robustness and real-time performance.