Practical Considerations about a 2d Algorithm for Object Orientation in Industrial Applications

Wr ht-re prese~it a sollition to a common prt>I)lrm in IMY. i.~, to identify and estimate the ori~ntation of touching m~clianiral pal ts on a planr surface. Aftrr an ini tid thr~sholding step, wp lisp x dist.ance tra~~sfnrnlation to crcatp a distanc~ map. 1Vr separate tli~ oI>jerts hy using wrrtprshpd segmentation nn the tlistmre map. Some ohjrrts may he s~grn~nterl into SPVP~R! parts. For pvpry segm~nt and every hnle we rrllrlll~k~ the cv1h-r r~f ~ravity for its surrou~ic!ine;cdge pixels. By this we haw nn st imxtion of thp orientation t~f thc objects. For thr r,l~jrcts that ronsist of only one APRI~PI~~, without hol~s, wc constrl~r.t x cirrle mund tllr rrr~tr~ of gravity. IP plot the ~pllups nf t11~ distance IIlnp or1 this circlr linp at a filnrtion of t,he anglr t~lltl itlcntify its maxi~iia ant1 nii~iirna. For tlw CIVP~ILII ro~~trtrl ~lgoritlim wc ilsr f~rzzy logics. To verify the irlc~ltifiration and to g~t il l~pt ter ~sti11ixtiorr of the orieritatini~ of the ohj~rt~, we finally do an edge niatchi~~~ HS~FIF; th~ distall~~ map which gives 11s q~~antit~ative rnetmur~m~nt~, of how well, the A~PS rnatrl~. This will givr us rnorp acrtirxt~ ~st~imations tI1a11 ran he acl~i~wd by ~"ttiistic~rl methods. Lal~or~rtory trsts llavr shown t,fiat nnr algorit,hnl will perform quit^ wcll or1 tllr shop-flour 1~1drr ccttain rest,rirtions ahot~t the (tist ~IICP between t Ilr carrlern nnrl the ohj~cts. MATCHING Edge matching : In order to clcterrninr. t hr c)ri~nt;rt,ion of thjerts WP want to nlxtch the edges of an nhjcct in xi1 inla~e with tllr CCIEPS. of the samv ohjert in a referrnr~ i~i~agt~~, IVe dn that by translating an11 rc>t.atirrg nne of th~rn until it.s rrlf;rb ~~XCIH mntrh thr ~rl~;r pixels of t,ll~ ntll~r ima~r RT wcll as possible. .4s mras~rr~m~nts of tlw difitrnrlr~s h~twrcr~ the two sets of PC~~P ~~XPIA IVP llsr tl~r ~rwthorl rpportpd by Barrow et a1 '. Ona of tlw e(l~p images is first transform~d into rr distanr~ Inap with some distance trslrlsforniation, i.r. tli~ rllarnfer nlgni ibllrrl, SPP Borgefors '. To ronipute the clistanre transformntiorl fur an in~xrp wit,ll edges, wr ralrulate the rlistm~c~ from CVP~Y pix~1 to the nearest. prlgr 11ix~1. Let rls call th~ list, of PCIEP ~IxPE coottlinatt.~ wc. g~t. from trarlsfr>rining and rotxtiiig them tht. Pattern Set. We 111.fii1~ mi Edge Matching Function (EMF), which we usv to Irlramrr tlw si~~~ilxrit>' \,r.twrrrl two P lr to ilrhirvp. If wr had a ronti~izr071s fi~nct.ion to mini~rlizr inst~i~11, tlir mlur of AEMF woldtl decrease, when iipproachin~ a rr~iniri~tirli. I11 011s CMP it will rlecrcxsr zlritil it is rlos~ to th~ ~ninirrlilnl, when suddpnlv it will inrs~xw. Tl~i.; is rlu~ to tl~r rhmgp in shnpe of the Pattcrri Set 011 translation arid / or rotation. IlFe can usr that ;w x sign to stop t11r iteration, heciiuw from that pc~jr~t t hp \)~Irxvitnlr of t llc itwittion is hard t,o predirt . Our algorithm shortrris t hr r-clml)~ltrrtin~~ time c-orl