Anew3Dsurfacecontouringandrangingsystembasedondigitalfringeprojectionandphaseshiftingtechniqueispresented.Usingthephase-shifttechnique,pointscloudwithhighspatialresolutionandlimitedaccuracycanbegenerated.Stereo-pairimagesobtainedfromtwocamerascanbeusedtocompute3Dworldcoordinatesofapointusingtraditionalactivetriangulationapproach,yetthecameracalibrationiscrucial.Neuralnetworkisawell-knownapproachtoapproximateanonlinearsystemwithoutanexplicitphysicalmodel,inthisworkitisusedtotrainthestereovisionapplicationsystemtocalculating3Dworldcoordinatessuchthatthecameracalibrationcanbebypassed.Thetrainingsetforneuralnetworkconsistsofavarietyofstereo-pairimagesandthecorresponding3Dworldcoordinates.Thepictureelementscorrespondenceproblemissolvedbyusingprojectedcolor-codedfringeswithdifferentorientations.Colorimbalanceiscompletelyeliminatedbythenewcolor-codedmethod.Oncethehighaccuracycorrespondenceof2Dimageswith3Dpointsisacquired,highprecision3Dpointscloudcanberecognizedbythewelltrainednet.Theobviousadvantageofthisapproachisthathighspatialresolutioncanbeobtainedbythephase-shiftingtechniqueandhighaccuracy3Dobjectpointcoordinatesareachievedbythewelltrainednetwhichisindependentofthecameramodelworksforanytypeofcamera.Someexperimentsverifiedtheperformanceofthemethod.
