简介：Inaerialrobots’visualnavigation,itisessentialyetverydiffculttodetecttheattitudeandpositionoftherobotsoperatedinrealtime.Byintroducinganewparametricmodel,theproblemcanbereducedfromalmostunmanageabletobepartlysolved,thoughnotfully,aspertherequirement.Inthisparametricapproach,amulti-scaleleastsquaremethodisformulatedfirst.Bypropagatingaswellasimprovingtheparametersdownfromlayertolayeroftheimagepyramid,anewglobalfeaturelinecanthenbedetectedtoparameterizetheattitudeoftherobots.Furthermore,thisapproachpavesthewayforsegmentingtheimageintodistinctparts,whichcanberealizedbydeployingaBayesianclassifieronthepicturecelllevel.ComparisonwiththeHoughtransformbasedmethodintermsofrobustnessandprecisionshowsthatthismulti-scaleleastsquarealgorithmisconsiderablymorerobusttonoises.Somediscussionsarealsogiven.