Finding Mean Traffic Speed in Low Frame-Rate Video

of 17

Please download to get full document.

View again

All materials on our website are shared by users. If you have any questions about copyright issues, please report us to resolve them. We are always happy to assist you.
17 pages
0 downs
Finding Mean Traffic Speed in Low Frame-rate Video Jared Friedman January 14, 2005 Final Project Report Computer Science 283 1 Abstract In this paper, I present a novel approach to estimate mean traffic speed using low frame-rate video taken from an uncalibrated camera. This approach takes advantage of a known relationship between traffic speed and traffic density to make tracking of individual vehicles unnecessary. The algorithm has been developed especially for nighttime conditions, though
  Finding Mean Traffic Speedin Low Frame-rate Video Jared Friedman January 14, 2005Fina !ro ect #eport$omputer Science 2%& 1   Abstract  'n t(i) paper, ' pre)ent a no*e approac( to e)timate mean traffic )peed u)ing owframe-rate *ideo ta+en from an uncairated camera T(i) approac( ta+e) ad*antage of a+nown reation)(ip etween traffic )peed and traffic den)ity to ma+e trac+ing of indi*idua *e(ice) unnece))ary T(e agorit(m (a) een de*eoped e)peciay for nig(ttime condition), t(oug( e.ten)ion) to daytime image) )eem /uite po))ie 't (a) een te)ted on )e*era image )e/uence) and )(own to produce re)ut) con)i)tent wit((uman e)timation) from t(o)e )e/uence)  Introduction $omputer *i)ion tec(ni/ue) (a*e een appied to image) of traffic )cene) for a*ariety of purpo)e) 1 ne of t(e more popuar of t(e)e purpo)e) i) to attempt to e.tracta mea)ure of t(e e*e of conge)tion of t(e road in t(e )cene !ropery di))eminated, t(i)information can e u)ed y dri*er) to pan route) t(at a*oid traffic and y fir)t re)ponder)to identify accident) T(e mea)urement typicay u)ed for conge)tion i) t(e mean )peedof traffic, a) t(i) i) t(e mea)urement a rationa tra*eer )(oud care aout 'n t(i) paper, ' pre)ent a new agorit(m to e)timate mean traffic )peed u)ing *ideo image) at ow frame)peed T(e wor+ i) moti*ated y t(e pre)ence of Trafficand, a new company t(at offer)*ideo )tream) from o*er 400 traffic camera) in t(e 3a)(ington $ area t(roug( freeinternet acce)) at wwwtrafficandcom6 !re*iou) wor+ on finding traffic )peed (a)wor+ed y finding it directy, e))entiay y trac+ing *e(ice) for a +nown time o*er a+nown di)tance and cacuating an a*erage ratio 7owe*er, due to andwidt( imitation),Trafficand8) camera) gi*e *ideo at e)) t(an one frame per )econd, wit( unreiae anddifficut to determine time inter*a) etween frame), ma+ing trac+ing e.tremey difficutif at a po))ie Some pui)(ed agorit(m)   2,& in)tead pace two *irtua ine) or 9tripwire): on t(e road at a +nown )eparation and mea)ure t(e time inter*a etween car)cro))ing t(e fir)t and cro))ing t(e )econd, in t(e natura computer *i)ion anaogue of t(e p(y)ica oop detector) on road) 3(ie t(i) may )eem to e a different tec(ni/ue fromtrac+ing, it )(are) many )imiaritie), incuding t(e a))umption t(at car) wi not tra*emuc( from one frame to t(e ne.t, and in practice it re/uire) an e*en (ig(er frame )peedt(an trac+ing ;e*ert(ee)), ceary (uman) are ae to udge traffic e*e) from Trafficand8)*ideo, and t(ey woud )ti e ae to do )o e*en if )(own ony e*ery two or t(ree frame),ma+ing trac+ing iteray impo))ie ' a))ert t(at t(e way we ma+e t(i) udgment i) ydetermining (ow co)ey )paced t(e car) are and u)ing t(e intuiti*e fact t(at co)ey)paced car) tend to tra*e more )owy t(an tig(ty pac+ed one) More preci)ey, we u)et(e in*er)e correation etween mean traffic )peed and traffic density , w(ic( i) defined a)*e(ice) per ane-mie 4 T(e approac( of t(i) paper i) to ta+e ad*antage of t(i)reation)(ip and compute den)ity directy, w(ic( i) ea)ier to compute at ow frame)peed), and con*ert t(i) into a )peed u)ing t(e +nown reation)(ip To compute den)ity in a particuar region of intere)t, we mu)t +now t(e numer of ane) of t(e region, t(e engt( of t(e region, and t(e numer of car) in t(e region in eac( 2  frame Some traffic *i)ion )y)tem) (a*e (ad to accommodate camera) t(at coud erotated and <oomed y traffic operator) wit( oy)tic+) eg, 1, 56, and t(u) (a*e (ad to uid in )ome automatic cairation capaiity to t(eir program) 7owe*er, Trafficand8)camera) appear to e )tationary, and we ma+e t(e a))umption t(at t(e numer of ane)and t(e engt( of t(e region need ony e determined once, and ta+e ad*antage of (umaninput in t(i) one-time ow-co)t )etup procedure Many pui)(ed and commercia)y)tem) = > a)o re/uire )ome initia (uman )etup? for one, if t(ere are mutipe road)or two direction) of a )inge road in t(e picture a) i) u)uay t(e ca)e6, t(e )oftwarecannot po))iy +now w(ic( road i) t(e intended one wit(out )ome (uman inputSpecificay, t(e initia cairation )etup )impy re/uire) a (uman to draw a rectange inword coordinate)6 around an area of intere)t and t(en to trace out t(e ane) in t(e region@)ing )ome )impe geometric con)traint) and a))uming a typica ane widt(, t(e engt( of t(e region can t(en e cacuated T(e cacuation of t(e numer of *e(ice) in t(e region pro*ed to e morec(aenging t(an e.pected T(i) partiay ecau)e )urpri)ingy itte pre*iou) wor+ (a) een done on t(e proem Se*era agorit(m) (a*e de*eoped e.ceent trac+ing of car)in daytime condition) at (ig( frame )peed), w(ic( impie) t(at t(ey are ae to recogni<e*e(ice) to )ome e.tent 7owe*er, in trac+ing *e(ice) directy, it i) not nece))ary to)egment car) propery, ut ony to identify o) t(at corre)pond to mutipe *e(ice) or  part) of *e(ice), )ince in a traffic )tream a t(e *e(ice), t(eir part) and t(eir )(adow)tend to mo*e at aout t(e )ame *eocity T(e agorit(m reported in 1 doe) re/uirecorrect )egmentation of *e(ice), ecau)e it mu)t e)timate t(eir )i<e correcty, ut itre/uire) correct )egmentation of ony a few *e(ice) at a time, and t(u) it )impy t(row)away any o) t(at do not corre)pond to a tig(ty defined *e(ice profie Accuratecounting of *e(ice) in daytime condition) re/uire) a more )op(i)ticated approac( to deawit( occu)ion, )(adow), and *e(ice) of widey *arying appearance 'n t(i) preiminaryreport, ' c(o)e to focu) on nig(ttime condition) ony and to ea*e daytime condition) for future wor+ ;ig(ttime condition) are ea)ier ecau)e at nig(t, it i) u)uay po))ie to)impy count t(e numer of (eadig(t) appearing in t(e region, and (eadig(t) are muc(more *i)ie and e)) *unerae to occu)ion, )(adow, and *arying appearance t(an car) ;ig(ttime condition) are in any ca)e a more )uitae potentia u)e of t(e agorit(mad*anced in t(i) paper, )ince trac+ing-a)ed )y)tem) ordinariy find daytime condition)muc( ea)ier t(an nig(ttime condition), gi*ing a t(e den)ity approac( a particuar ad*antage in t(e)e condition)T(i) paper fir)t re*iew) t(e +ey a))umption) of t(e agorit(m and di)cu))e) t(eir *aidity and w(ic( one) coud e rea.ed in furt(er wor+ ' t(en di)cu)) in detai t(ewor+ing) of t(e agorit(m and foow y )ome con)ideration) of it) computationaefficiency ' concude wit( empirica re)ut) *aidating t(e accuracy of t(e agorit(m Underlying Assumptions T(e foowing gi*e) a i)t of t(e +ey a))umption) u)ed to )impify t(e proemand )ome di)cu))ion of t(eir *aidity &  16 'mage) are ta+en at nig(t 7eadig(t) are t(e rig(te)t o ect) in t(e region of intere)t26 Traffic i) mo*ing generay toward) t(e camera, ut not amo)t6 directy into it T(e)econd re/uirement e.i)t) ecau)e w(en traffic i) going amo)t directy toward) t(ecamera, t(e gare from t(e (eadig(t) create) oom, en) fare), and )e*ere di)tortionT(e fir)t re/uirement e.i)t) ecau)e if t(e traffic i) mo*ing away from t(e camera, t(e(eadig(t) wi not e directy *i)ie 'n my opinion, t(e )econd of t(e)e twinre/uirement) i) muc( more rea)onae t(an t(e fir)t Many of Trafficand8) nig(ttimeimage) are )o )e*erey di)torted y t(e en) fare) t(at it i) neary impo))ie e*en for a(uman to determine t(e amount of traffic, and wor+ing wit( t(e)e image) woud e a reac(aenge 7owe*er, trac+ing car) going away from t(e camera wi o*iou)y enece))ary for a fieded )y)tem, and future )y)tem) coud u)e eit(er t(e rear *e(ice ig(t)or t(e fairy rig(t refected gare from t(e (eadig(t) to accompi)( t(i)&6 Ve(ice) are confined to t(e road pane, and t(ere e.i)t) a region of intere)t wit()traig(t edge) A)o, t(e numer of ane) in t(e region of intere)t i) con)tant T(e)ere/uirement) are nece))ary for t(e cacuation of t(e geometry of t(e )ituation46 T(e widt( of eac( ane in t(e picture i) appro.imatey 115 feet T(i) a))umption i)t(e di)tance re/uired to determine t(e )cae of t(e image and t(u) t(e engt( of t(e regionof intere)t T(e *aidity of t(e a))umption i) ta+en from % w(ic( )tate) t(at *irtuay aAmerican (ig(way) (a*e ane widt() etween 10 and 125 feet at a time), wit( anewidt() co)e to 12 feet eing t(e mo)t common t(er )y)tem) (a*e u)ed a *ariety of mean) to attempt to produce a )cae mea)urement, mo)ty y pacing p(y)ica mar+) ont(e road   B, 10, at(oug( 1 did )o y a))uming a +nown di)triution of *e(iceengt() 7owe*er, in t(i) )ituation it wa) impo))ie to (a*e an operator pacing mar+)on or near t(e road C)timating y mean *e(ice engt( re/uire) an agorit(m t(at canaccuratey determine *e(ice engt( of a *e(ice), incuding truc+), w(ic( i) difficut tocon)truct and mu)t e run o*er a con)iderae period to get an accurate mean *aueFurt(ermore, it i) not at a cear t(at t(i) mean *e(ice engt( i) more con)tant from roadto road t(an t(e mean ane widt( 11 report) e*idence t(at t(e mean *e(ice engt(c(anged con)ideray depending on t(e time of day, t(e (ig(way, and t(e ane o)er*ed, primariy due to t(e con)iderae *ariation in t(e pre)ence of arge truc+), eading to argeerror) in )y)tem) t(at a))umed a con)tant *e(ice engt( @)ing t(e ane widt( a) acairation too appear) to e a no*e )ugge)tion, and it )eem) a )en)ie c(oice for a*ariety of )ituation), not imited to ow-frame rate *ideo 't i) per(ap) wort( noting t(atif t(e ane widt( did not (od to t(e 10-125 foot range, t(en t(e *aidity of a))umption56 woud e in /ue)tion anyway, a) t(i) woud affect t(e den)ity-)peed reation)(ip56 Traffic )peed and den)ity (a*e a +nown, and con)tant reation)(ip, )pecificay Cdie8)mode a) gi*en in 4 T(i) a))umption i) admittedy )omew(at contro*er)ia 3(ie t(ein*er)e correation etween )peed and den)ity i) o*iou), t(e e.act reation)(ip (a) eena topic of con)iderae deate For decade), it wa) eie*ed to e a inear reation)(ip ont(e a)i) of a )inge )tudy u)ing )e*en data point) a coected from a )inge (ig(way 4Furt(er )tudy y Dreenerg found t(at a ogarit(mic reation)(ip wa) t(e e)t fit, a) inFig 1 7owe*er, de)pite t(e )eemingy e.ceent fit, a numer of ca*eat) can e rai)ed 4
Related Search
We Need Your Support
Thank you for visiting our website and your interest in our free products and services. We are nonprofit website to share and download documents. To the running of this website, we need your help to support us.

Thanks to everyone for your continued support.

No, Thanks