基于视觉伺服的小型四旋翼无人机自主飞行控制研究进展吕强1,马建业1,王国胜1,林辉灿1,梁冰2科技导报2016,34(24)无人机平台的优势:•无人机正快速成为一种无处不在的现代工具,它在搜索、救援、监视、航拍、远程遥感等领域已得到广泛应用,更被应用于电线检修和农业感知等特殊领域。与有人驾驶的飞行器和固定翼无人机相比,小型四旋翼无人机由于体型小、可垂直起降、机动灵敏等特点,使针对狭窄空间的搜索成为可能,并且在室内进行飞行测试更加方便;此外其以电池替代燃料驱动,即使发生碰撞也不会对人类造成严重威胁,安全性得到大幅提升。视觉伺服简介:•视觉伺服指的是利用计算机视觉信息控制机器人的运动,依赖于图像处理、计算机视觉和控制理论等技术。•在GPS信号较弱甚至失效的环境下,视觉伺服能够通过视觉信息控制自主飞行,因此近年来视觉伺服在自主飞行控制领域受到广泛关注。对于小型四旋翼无人机自主飞行控制的应用研究中,视觉伺服的实时性、精确性和鲁棒性尚待提高,且小型四旋翼无人机的智能化不高,在室内室外模式转换及室内协同控制方面还有广阔的发展空间。视觉伺服的分类:•所有基于视觉伺服的方案,目的都是减小误差e(t)。误差e(t)可表示为•e(t)=s(m(t),a)-s*•式中,矢量m(t)为一连串的图像测量值;s(m(t),a)为视觉特征点矢量;a为摄像头固有参数;s*为期望特征值。•通过对s定义的不同,可将视觉伺服分为基于位置的视觉伺服(PositionBasedVisualServo,PBVS)和基于图像的视觉伺服(Image-BasedVisualServo,IBVS)。PBVS的基本结构优点:可以直观地在直角坐标空间定义目标的运动,符合现有机器人的工作方式。缺点:控制精度很大程度上依赖于位姿估计精度,而位姿估计精度依赖于摄像机和机器人的标定精度等;此外计算量较大。IBVS的基本结构优点:无需三维空间定位、对摄像机和机器人标定不敏感;计算量较小。缺点:伺服控制器复杂且缺乏适应性;需要额外的传感器获取深度信息;移位过大会导致不可预知摄像头运动。视觉伺服无人机发展趋势:•1)视觉算法的鲁棒性。若摄像头运动过快且特征不是特别明显,则会发生特征点跟踪失败从而丢失位置的现象。只有从根本上增强视觉算法的鲁棒性,才能够实现在任何环境下的自主飞行。•2)室内室外模式转换。当GPS信号低于一定值时转换为视觉定位,高于一定值时转换为GPS定位。•3)室内协同控制。多无人机室内协同控制可同时控制多架无人机对复杂环境进行分块搜索、侦察和构图等,这一功能不仅大大提高任务执行效率,而且增强其可靠性。MonocularVision-BasedAutonomousNavigationSystemonaToyQuadcopterinUnknownEnvironmentsRuiHuang1,PingTan1,BenM.Chen2InternationalConferenceonUnmannedAircraftSystems2015(ICUAS2015).NewYork,USA:InstituteofElectricalandElectronicEngineers(IEEE),2015:1260-1269.VideoDemoI:SquarePathFollowingVideoDemoII:CirclePathFollowingVideoDemoIII:Hovering.PlatformTheAR.DroneisamicroUnmannedAerialVehicle(UAV)developedbytheParrotcompanyfortheincreasingmarketofvideosgamesandhomeentertainment.Thedroneislowcost,safe,andeasytoflybyendusers.A.Hardware1.afrontcamera2.downward-lookingcamera3.Thenavigationboardisequippedwiththesensorsincludinga3-axisaccelerometers,a2-aixsgyroscope,a1-axisverticalgyroscopeand2ultrosonicsensors.•B.PlatformModelingandControllerDesignWeusetheSystemIdentificationToolboxofMATLABtofitthesystemtransferfunctionsforeachchannelseparately.Firstly,wesampleasequenceofachirpsignalasthecontrolparameters.ThenwesendthecontrolsignalsatthesamplingratetothedroneandrecordtheresponseusingamotioncapturesystemcalledViconwithmillimeterprecision..VisualSlamA.FeatureTrackingB.CameraPoseEstimationC.KeyframeInsertionandMapGenerationD.RelocalisationSENSORFUSIONAdvantage:•Firstly,weuseamuchsimplifiedstatespacetoreducethecomputation.•Secondly,ourscaleestimatordoesnotrelymuchonverticalmotionofthedrone,whichisnotcommonintherealflights.•ToaddressthedriftingerrorsinintegratingIMUmeasurements,wesolvethescalefactorbycomparingvelocitiesfromvisualSLAMandonboardsensors.总结:•该系统将通过无线摄像头获取的视频流传输给地面站计算机,在地面站上运行视觉SLAM算法处理数据进行位姿估计。为解决收到的视频中运动模糊和帧丢失问题,改进了视觉SLAM算法中的特征跟踪方法和重新定位模块,增强了鲁棒性。同时为获得更精准的三维位置与速度,融合了视觉SLAM和机载传感器获取的数据,并且设计了一种扩展卡尔曼滤波器用于传感器数据融合,矫正了局部漂移误差并且解决了比例模糊问题。AutonomousFlightandObstacleAvoidanceofaQuadrotorByMonocularSLAMOmidEsrafilianandHamidDIn20164thInternationalConferenceonRoboticsandMechatronics(ICROM),pages240–245.AutonomousFlightandObstacleAvoidanceofaQuadrotorByMonocularSLAMMAPENRICHMENTFirst,thepointcloudareaofrobotisselectedandisdividedintoseveralboxesinordertofitaplanetoeachbox.Then,theseboxesaredistinguishedindifferentclusters(Fig.3-b).Inthenextsteptheoutlierplanesaredeterminedandcorrectedaccordingtothenormalvectorofeachcluster(Fig.3-c).Finally,theexistedpointsarelinedupandsomeextrapointsareaddedtomakeitmoredense(greenpointsinFig.3-d).PATHPLANNINGASimpleAutonomousFlightControlMethodofQuadrotorHelicopterusingonlysingleWebCamera*KazuyaSato1,ToruKasahara1,andTomoyukiIzu2InternationalConferenceonUnmannedAircraftSystems2016(ICUAS)2016TwocoloredballsasacolormarkerareattachedtothemulticopterandrecognizethepositionofcoloredballsusingcapturedvideoimagesonthewindowfromaWebcamerawhichputtingontheground.Thecapturedvideoimageoftheactualflightexperiments.EXPERIMENTALRESULT•Thepositioninformationofmulticopterinx,y,zcoordinatesweremeasuredbyusingcolormarkerswithsingleWebcamera•Caneasilymanipulateamulticopteronlyoperatingthechangingthewidthoftheframeonthewindow.•Doesnotneedtomodifytheactualmulticopterwhichismanipulatedbythehumanoperatorusingaradiocontroltransmitter.