Research On Visual SLAM’s Back-End Methods Of UAV Based On Filtering Algorithm

With the development of the control technology in unmanned aerial vehicles(UAV)field,the application of UAVs has a broad space in both civil and military fields and this trend has emerged.In order to achieve higher intelligence and autonomy of UAVs,which is an example in completing a task accurately without human control,the primary problem to solve is autonomous navigation,location and map building.Applied the SLAM technology into UAV navigation is the research focus.This paper mainly studies the Back-End optimization algorithms of UAV visual SLAM.These algorithms can locate the UAV using the visual sensor information.With the addition of the observation information of feature points,the position of the UAV and feature points can be updated continuously,and it makes the UAV complete the autonomous flight and navigation mission.Firstly,this paper introduces the disadvantages of UAVs’ navigation scheme and summarizes the merit of the quadrotor UAVs.Next,it demonstrates the Importance and necessity of UAVs realizing the capability of simultaneous localization and mapping(SLAM).Then this paper presents the classical visual SLAM framework and the research difficulties of SALM on UAVs,and explains the main research contents.Secondly,by building the dynamic model of the quadrotor UAV,this paper presents the state equation and observation equation of UAV visual SLAM.And this is the basics of the following research.Thirdly,in order to use filter optimization algorithms in the SLAM,the posterior probability of state variables in the SLAM problem is expressed as a form of recursive solution with known variables in this paper.Then the Extended-Kalman-Filtering-based SLAM(EKF-SLAM)algorithm is studied.Aiming at the defects of EKF-SLAM,such as existing linearization error and large amount of calculation,this paper studies the SLAM algorithm based on Rao-Blackwellized Particle Filter(FastSLAM).The FastSLAM algorithm inherits the implementation and nonlinearity of particle filter,and its Operation speed is faster than EKF-SLAM.But there are still some disadvantages existing.Fourthly,this paper studies the disadvantages of the FastSLAM in detail.Aiming at these disadvantages,Particle Swarm Optimization(PSO)algorithm and Artificial Fish Swarm algorithm(AFSA)are combined with FastSLAM respectively in this paper and these two methods surmount the disadvantages of particle filter,such as inferior-best importance distribution,particle degradation and sample exhaustion.In the end,according to the development trend of UAV’s visual SLAM technology in recent years,the follow-up research work of this paper is present.