Research Of Vision SLAM Based On Unmanned Aerial Vehicle

Recently,with the rapid development of computer control technology,computer vision technology and artificial intelligence technology,which not only makes unmanned aerial vehicle(UAV)from the traditional military field gradually into all aspects of the civil field,but also in the aspect of the operation mode of the UAV changes greatly,that is to say,from the traditional way of remote manual control before gradually transition to now fully customized control.As a result of the UAV intelligent and customized development cannot leave the navigation and positioning technology,the traditional navigation system mainly uses the GPS,and when UAV does tasks in some special environment,in which it is vulnerable to the interference of environment,and the GPS signal will become weak or complete failure.In order to make up for the lack of GPS navigation system,the SLAM technology is used in this paper for being an auxiliary navigation way of UAV.The UAV can realize the establishment of environment map and uses it for location without external signal by SLAM.This topic research is based on the small UAV monocular visual SLAM technology,because the UAV in the actual process of flight is not a kind of smooth movement,so it can't adopt the way of inputting control command to predict the position of the UAV in the next moment.Therefore,this paper uses the visual odometer to estimate the position of the UAV in the next moment.In the process of feature points extraction of visual odometer requires very large amount of calculation,the high real-time ORB is proposed as the image feature extraction algotithm in this thesis,and speeds up the image feature extraction.As to the optimization algorithm of SLAM system,this paper chooses the sparse extended information filter,due to its time complexity and space complexity are better than that of extended kalman filtering algorithm;In the SLAM system based on UAV,the real-time,UAV pose estimation accuracy and consistency of algorithm has higher requirements;so this paper has proposed two kinds of improved algorithm based on the basic algotithm.Firstly,from the perspective of sparse operation,and using the entropy properties,integrated the current and the next observation to choose the sparse feature point which is the weakest correlation to posture,so as to effectively improve the accuracy and consistency of the algorithm.And then according to sparse distribution characteristics of information matrix elements,the tridiagonal matrix is used to quickly solve inverse operation in the state average recovery process,thus effectively improve the computation efficiency.Finally,establishing the simulation model of 3d environment,which is composed of the pre-set flight path of UAV and environmental feature points,and using the improved SEIF SLAM and the standard SEIF SLAM to the simulation environment for verification and comparing with the simulation results.the simulation results show,the improved algorithm that is based on the entropy properties in the pose estimation precision and consistency compared with the standard algorithm has obvious improvement;the improved algorithm that is based on the tridiagonal matrix inversion in the computation efficiency is faster than the standard algorithm as the order number of information matrix increased gradually.