Abstract metric to nd the optimal pose and to measure the distance between the measurements

Size: px
Start display at page:

Download "Abstract metric to nd the optimal pose and to measure the distance between the measurements"

Transcription

1 3D Dstance Metrc for Pose Estmaton and Object Recognton from 2D Projectons Yacov Hel-Or The Wezmann Insttute of Scence Dept. of Appled Mathematcs and Computer Scence Rehovot 761, ISRAEL Abstract Model based object recognton and model based pose estmaton requre a dstance metrc to nd the optmal pose and to measure the dstance between the measurements and possble models durng the recognton process. When the measurements are gven n 2D (such as n orthographc and perspectve projectons) the commonly used dstance between the 3D model features and the 2D mage features s the 2D Eucldean dstance measured n the mage plane. However, ths 2D dstance does not, usually, ncrease monotoncally wth the real 3D dstance and thus does not really represent the dstance beng measured. In ths paper we propose a new scheme n whch both the optmal postonng and the evaluaton of smlarty between the 2D mage and the model s performed relatve to the 3D dstance. Ths dstance s calculated between the model features and a 3D predcted object whch s a permssble reconstructon of the measured object and s the \closest" to the model features. 1

2 1 Introducton Model based object recognton and model based pose estmaton are two complementary problems that arse frequently n the vson lterature (for example [8, 9, 5, 4, 12, 6]). In model-based pose estmaton the poston n 3D space (translaton + orentaton) of a known object s determned from the object measurements. In model-based recognton a measured object s compared to a lbrary of prototype models n order to nd the model whch sthe \closest" to the vewed object. Commonly, ths process requres the estmaton of the best model postonng followed by a measure of smlartybetween the model and the measured data. Both the recognton process and the pose estmaton process, use some dstance metrc to nd the optmal pose and later to measure the dstance between the measurements and the hypothess model. Many applcatons use feature ponts such as maxmum curvature, segmentendponts and corners as a model descrpton. In these applcatons, when measurements are gven n 3D (such as range nder or stereo data), Eucldean dstance between the 3D model features and the correspondng 3D measured features s calculated to gude the optmal pose estmaton and the recognton process [11, 8, 5, 2]. If fx g =1:::n and fx g =1:::n are two setsof3d vectors representng the locatons of the model-ponts and the measured ponts respectvely, then the 3D Eucldean dstance between these two setss: D 3 = X kt (X ) ; X k 2 where T s the rgd transformaton representng the model pose. Ths dstance metrc s reasonable snce t descrbes the amount of3d dstorton a model has to be undergo n order to t the measurements. In model-based pose estmaton we am to mnmze ths dstance subject to the transformaton T and n recognton tasks the mnmzaton s performed subject to T and the model fx g. More frequently, recognton tasks deal wth measured features gven n 2D (projected mages). In ths case a commonly used dstance between the 3D model features and the 2D mage features s the 2D Eucldean dstance measured n the mage plane. If fm g =1:::n are 2D vectors representng the locatons of n correspondng measurements then the 2D Eucldean dstance s dened as: D 2 = X kt (X ) ; m k 2 2

3 where s a projecton operator. In ths case, the measured features n the mage are compared wth the model features projected onto the mage plane. The mnmzaton of ths 2D Eucldean dstance s used to gude the postonng and the recognton processes [12, 3, 14, 15, 7]. However, the problem wth ths 2D dstance s that n most cases, ths dstance does not ncrease monotoncally wth the real 3D dstance and thus does not really represent the 3D dstortons the model must undergo n order to t the measured object. Ths nadequacy of the 2D dstance causes mprecsons n the recognton and postonng especally when dealng wth perspectve projecton. For example, n perspectve mages, a 2D dstance between a measured pont and ts assocated projected model pont canhave varyng values n the real 3D dstance, accordng to the depth of these ponts n 3D. Therefore, the relatve contrbuton of ths par n the total 2D dstance may der from ts contrbuton n the total 3D dstance. The same knd of problem arses when dealng wth perturbatons n the measurements or n the model features. These perturbatons nuence derently the 2D dstance and the 3D dstance. In ths paper we propose a new scheme whch, gven a 2D mage and a canddate model, calculates the optmal postonng of the model and evaluates the \smlarty" between the 2D mage and the model. Both, the optmal postonng and the evaluaton of smlarty s performed relatve to the 3D dstance. Ths dstance s calculated between the model features and a 3D predcted object whch s a permssble reconstructon of the measured object. More precsely, we calculate the mnmum dstorton of the model whch produces a permssble reconstructon of the measured features, where a reconstructed object s consdered permssble f t does not contradct the measured data. The permssble reconstructed object whch s the \closest" to the canddate model (n terms of mnmum dstorton) s the predcted object. In ths paper we descrbe a method that deals wth recognton and pose-estmaton tasks based on 3D dstance metrc. In ths scheme we nclude uncertanty both, n the measurements and n the model features. We show that the superorty of our scheme s manly when the uncertanty n the model s sgncant. Ths characterstc becomes mportant when modelng non-rgd objects or when dealng wth uncertan models whch are used to mprove recognton of a partcular nstance of a class dened by a general model. The rest of ths paper s organzed as follows: Sectons 1 and 2 propose the general framework for dealng wth nose free measurements and precse model denton. Sectons 3

4 3,4, and 5 extend the proposed framework to deal wth uncertan models and nosy measurements. Sectons 6 and 7 present the process for pose estmaton and recognton. Some smulated results are gven n Secton 8. 2 Dentons A model of a 3D object s represented by asetofponts: fx g =1n where X =(x y z ) t s assocated wth the locaton of the th model pont and represented n an object-centered frame of reference (x y z ). A transformed model s the coordnates of the model ponts as gven relatve to the vewercentered frame of reference (x y z). If T denotes the rgd transformaton between the object-centered frame and the vewer-centered frame then the transformed model s descrbed by the collecton: fx g =1n where X =(x y z ) t = T (X ). A measurement of a 3D object s a set: fm g =1n where m represents a measurement ofthe th feature pont of the object. Ths paper deals wth the case where the measurements are obtaned from a projecton of the object onto a 2D mage plane. Thus, m =(v w ) t s represented n the mage frame of reference (v w). A predcted object s a set of 3D coordnates of the form: fu g =1n where U s an estmate of the th feature pont of the object and represented n the vewercentered frame of reference (see Fgure 1). The predcted object s the \closest" object to any transformed model satsfyng the mage constrants. Formally, we choose such an object fu g whch mnmzes the followng quantty: C = nx kt(x ) ; U k 2 (1) 4

5 z z x y T(x ) U lne(m ) v w m f y x Fgure 1: General conguraton of a perspectve projecton. The frames (x y z ), (x y z) and(v w) represent the object-centered, the vewercentered and the mage frame of reference, respectvely. under the set of constrants: project(u )=m for each =1n (2) where project(u ) s the projecton of U onto the mage plane and where the transformaton T s any rgd transformaton. 3 The Predcted Object and the Optmal Transformaton Mnmzaton of C n Equaton 1 requres evaluaton of both, the optmal transformaton T and the \closest" predcted object fu g =1n. However, knowng the optmal transformaton, t s straghtforward to nd the predcted object. In the followng we elaborate on ndng the optmal transformaton T wth respect to any predcted object. Denote by lne(m ) the collecton of ponts n 3D space where the feature pont U can 5

6 be located.e. lne(m )=fv j project(v) =m g : If the measurement s an orthographc projecton onto the plane z = then lne(m ) s a lne parallel to the z axs and passng through the pont (m t ) t. If the measurement s a perspectve projecton, as depcted n Fgure 1, then ths lne passes through the focal pont ( ) and the pont (m t f)where f s the focal length. If Q s a pont n3d we denote by kq ; lne(m )k 2 the squared Eucldean dstance between lne(m ) and the pont Q. Lemma: If fu g s the optmal predcted object mnmzng Equaton 1, then: kt(x ) ; U k 2 = kt(x ) ; lne(m )k 2 for each =1n : Proof: By denton kt(x ) ; lne(m )k 2 = mn kt(x ) ; Vk 2 V2lne(m ) However, due to the mage constrant U 2 lne(m ). Snce U s chosen s.t. kt(x ) ; U k 2 wll be mnmzed, we have: kt(x ) ; U k 2 = mn kt(x ) ; Vk 2 V2lne(m ) and thus kt(x ) ; U k 2 = kt(x ) ; lne(m )k 2 : 2 From the lemma t s clear that the optmal transformaton ^T can be derved as follows: ( ^T =arg mn T nx kt (X ) ; lne(m )k 2 ) : After ^T s calculated, the second stage s to nd the predcted object. each feature pont s chosen so that Equaton 1 s mnmzed, thus: The locaton of ( ) U = arg mn k ^T(X ) ; Vk 2 V2lne(m ) for each =1n : 6

7 d z c b e f g a y x Fgure 2: Perspectve projecton of nose. Smlar nose n the model ponts (d and e) are projected derently onto the mage plane (f and g), dependng on the dstance from the focal pont. Conversely: nose n the mage plane (a) can have avaryng nuence on the predcted object, accordng to the depth of the predcted pont (b and c). 4 Uncertanty n the Model and n the Measurements In the prevous secton no knowledge about the measurement uncertantes and the model uncertantes was assumed. Snce we work n a nosy envronment, t s mportant tocharac- terze these uncertantes especally when the measurements are taken followng a perspectve projecton. Wth ths knd of projecton, smlar nose n the model ponts (d and e n Fgure 2) are projected derently onto the mage plane (f and g n Fgure 2), dependng on the dstance from the focal pont. And conversely: nose n the mage plane (a n Fgure 2) can have avaryng nuence on the predcted object, accordng to the depth of the predcted pont (b and c n Fgure 2). In ths secton we consder measurements and model ponts whch are assocated wth some uncertanty. That s, each measurement snow represented by apar: measurement()=(^m ) 7

8 where ^m s the actual measurement of a real value m s.t. ^m = m + ".We assume that the nose term " s of zero mean and ts covarance matrx s known. The covarance matrx depcts the uncertanty n the actual measurement and s manly due to two factors: Uncertanty due to measurement nose (e.g. dgtzaton, blurrng and chromatc aberratons). Uncertanty dependent upon the feature detecton process. For example, a detected end-pont of a lne segment wll have low postonal uncertanty n the drecton perpendcular to the lne segment and a hgh uncertanty n ts drecton. Smlar to the measurements, we assocate for each model-pont acovarance matrx that each one of the model ponts s represented by a par: so model-pont() =(^X ) where ^X s an estmated locaton of the th model pont wth a 3 3covarance matrx. Ths covarance matrx denotes the uncertanty of the estmated locaton ^X that may arse from three sources: Uncertanty due to mprecse modelng of the 3D object. Uncertanty due to modelng a class of 3D objects, for example, modelng a general face. Uncertanty due to modelng non rgd objects such as rubber objects. In some cases there s no knowledge about the nose n the measurements or n the model. In such cases we assocate an dentcal covarance matrx wth each measurement or model-pont suchthat = I or = I where I s the dentty matrx and are some scalars. When dealng wth an uncertan model and nosy measurements t s clear that t s erroneous to nd a predcted object whch mnmzes C of Equaton 1. In ths equaton all terms added n the rght-hand sde nuence the soluton equally where n our case the 8

9 nuence should be nversely proportonal to the uncertanty ofeach term. Therefore the predcted object has to be calculated subject to a Mahalanobs dstance: C = nx [(T (X ) ; U ) t W ;1 (T (X ) ; U )] (3) where W sa33 covarance representng the uncertantyof(t (X );U ). Ths uncertanty can be deduced from and as wll be explaned later on (Secton 7.1). It should be noted that even f we don't have any aprorknowledge about the uncertanty n the model and n the measurements so that all are dentcal and all are dentcal as well, the term C n Equaton 1 s ncorrect and the transformaton T should be found subject to mnmzng Equaton 3. Ths s true snce W n Equaton 3 depends also on the depth of each pont from the mage plane as can be seen n Fgure 2. 5 The Mahalanobs Dstance to an Uncertan Lne Accordng to the lemma gven n Secton 3, fu g n Equaton 3 can be replaced by flne(m )g. However, n ths case, these lnes are uncertan snce they are deduced from uncertan measurements. Therefore, we are actually nterested n the Mahalanobs dstance (M.D.) between the transformed model to these uncertan lnes. Ths secton elaborates the representaton of an uncertan lne and the measure of the M.D. to t. Let u be a 3 dmensonal vector of random varables where Efug = ^u and varfug = u. Efug denotes the expectaton value of u and varfug = Ef(u ; ^u) t (u ; ^u)g denotes ts 3 3 covarance matrx. If p s some pont n 3D space (wth zero uncertanty) then the squared Mahalanobs dstance between p and ^u s dened as: d 2 =(^u ; p) t ;1 u (^u ; p) Assume the covarance matrx u s dagonal, thus there s no correlaton between the components of u: u = 2 x 2 y 2 z If we set 2 z to be an nnte value then the contours of constant M.D. from ^u wll be ellptc cylnders parallel to the z axs centered at ^u (See Fgure 3). The cross secton of these 1 C A 9

10 (x,y) = (u ^,u ) ^ x y u^ constant Mahalanobs dstance z y x Fgure 3: The Mahalanobs Dstance to an uncertan lne. When the uncertan lne s parallel to the z axs (.e. the dagonal covarance matrx u assocated wth the lne has an nnte value for z)thenthecontours 2 of constant Mahalanobs Dstance wll be ellptc cylnders parallel to the z axs. The cross secton of these cylnders along a plane parallel to the (x y) plane s composed of enclosed ellpses havng length and breadth proportonal to 2 x and 2 y n the prncpal drectons. cylnders along a plane parallel to the (x y) plane s composed of enclosed ellpses havng length and breadth proportonal to 2 x and 2 y n the prncpal drectons. Therefore, n ths case, the M.D. between a pont p and ^u depends on the dstance and drecton of p from a lne perpendcular to the (x y) plane and passng through the pont ^u. Explctly ths dstance s gven by: d 2 =(^u ; p) t 1 2 x 1 2 y 1 C A (^u ; p) =(^u x ; p x ) 2 + (^u y ; p y ) 2 2 x 2 y We can vew ths dstance as a M.D. from an uncertan lne wth some uncertanty nthe drectons perpendcular to the lne drecton. Ths uncertanty s gven by the covarance matrx: lne = 2 x 2 y 1! : (4)

11 Note, that the M.D. between p to any pont on ths lne s equal to the M.D. between p and ^u as can be seen from Equaton 4. Ths proposton s not lmted to lnes parallel to one of the axes and s vald for any uncertan lne. For example, consder an uncertan lne passng through pont ^v wth drecton represented by the unt vector R^z where R s some rotaton matrx and ^z sauntvector algned wth the z axs. Assumng the uncertanty of ths lne s the same as the prevous one, ths lne can be represented by an uncertan pont (^v v ) where ts covarance matrx s as follows: v = R 2 x 2 y 1 1 C A R t The M.D. between pont p to any pont on ths lne s n fact the M.D. from p to the uncertan pont ^v. In the followng sectons we show how to use ths dea to nd the optmal transformaton of a model and to nd the predcted object closest to the model n the sense of 3D Eucldean dstance. The method descrbed below fuses the nformaton from all the measured ponts and estmates the model transformaton T by ncremental renements usng Kalman-f lter [13, 16]. At each step a new uncertan lne s generated from the assocated measurement and an updated soluton s produced. 6 Convertng the Measurements to Uncertan Lnes As stated n Secton 3 the optmal transformaton s calculated subject to mnmzng the dstance between the transformed model and the lnes flne( ^m )g. These lnes are n fact uncertan lnes due to the uncertanty n the measurements. Each of these lnes can be represented by a3d pont ^M located on lne( ^m )andhavng a covarance matrx ;. The pont uncertanty n the drecton of projecton s nnte, and ts uncertanty n the perpendcular drectons s deduced from the assocated measurement uncertanty. Ths uncertan 3D pont (uncertan lne), ( ^M ; ), can be consdered as a 3D measurement of the th transformed model-pont T ( ^X ). In ths secton we explan how to convert the measurements nto uncertan lnes and n the followng secton we gve the algorthm for ndng the optmal transformaton usng ths nformaton. 11

12 Assume that the measurements are performed on the mage plane usng the coordnate system (v w) where the k th measured pont s: measurement(k)=[^m k =(^v k ^w k ) k vw] k sa22covarance matrx descrbng the uncertanty of the actual measurement(^v vw k ^w k ). For smplcty we ommt the subscrpt k at ths tme so that ths measurement s represented by the par [(^v ^w) vw ]. We separate our dscusson nto two cases: orthographc projecton and perspectve projecton. Orthographc projecton: In the case where the projecton s along the z-axs (orthographc) the assocated uncertan lne s represented as: ( ^M ;) = [(^v ^w ^z) where ^z s any estmate of the z coordnate. vw 1! ] Perspectve projecton: In the case of perspectve projecton, the modelng of the uncertanty s more complex. Assume that the orgn of the vewer-centered frame of reference (x y z) s at the focal pont as shown n Fgure 4 and the focal length s equal to one. We am to transform the measurement ^m =(^v ^w) gven n the mage-plane coordnate system (v w) nto a representaton n the Cartesan system (x y z). Consderng the sphercal coordnate system (r ) (Fgure 4). The vector (^v ^w) determnes the angular coordnates ( ) but leaves the value of r undetermned: ^ = arctan( p^v 2 +^w 2 ) ^ = ^v arccos( p^v2 +^w ) : 2 Addtonally, the uncertanty of(^v ^w) s transformed nto a covarance matrx n the ( ) system as follows: =!! ) w) ) s the Jacoban of the transform from (v w) to( ), and the w) 12

13 z u r v f φ w m^ y θ x Fgure 4: A perspectve projecton of the pont U onto the mage plane (v w). The pont can be represented n ether, a Cartesan system (x y z) or a sphercal system (r ). at pont (^v ^w). The Jacoban matrx w) = ^v 2 ^w 2 ; ^w 2 ^v 2 where 1 p^v2 +^w p^v2 +^w : 2 The transformaton nto sphercal coordnates, as an ntermedary stage, allows a smple representaton of the assocated uncertan lne: ( ^M ;) = [(^r ^ ^) r ] where r = 1 1 C A and ^, ^, are the expressons descrbed above. ^r s unknown but an estmaton of ^r wll be chosen as s explaned later n ths secton. 13

14 In practce we arenterested n representng the uncertan lne n Cartesan coordnates, thus, the representaton s transformed agan from the sphercal coordnates to Cartesan coordnates (x y z) as follows: ( ^M ;)=[(^x ^y ^z) xyz ] where ^x = ^r sn ^ cos ^ =^r^v ^y = ^r sn ^ sn ^ =^r ^w ^z = ^r cos ^ =^r and the covarance matrx s: xyz =!! y y z) r ) The Jacoban y ) = ^v ^r^v ;^r ^w ^w ^r ^w ^r^v ;^r= 1 C A where the dervatve staken at the pont (^r ^ ^). Here too, all values are known except for ^r. Snce the process that estmates the optmal transformaton ncrementally mproves the k;1 estmaton of T,.e, at each step k, there exsts an estmate ^T from the prevous step, we use ths estmate to calculate an estmate of ^r at step k as follows: ^r k k;1 = k ^T ( ^X )k where ^X s the locaton of the correspondng pont n the model. We emphaszethatthe uncertanty of ths estmate, as expressed n the covarance matrx, s nnte. 7 Estmaton of the Optmal Transformaton As stated, the rst step n ndng the predcted object s to estmate a transformaton T whch optmally translates the ponts f( ^X k k)g of the model onto the correspondng uncertan lnes f( ^M k ; k )g, generated from the measurements f( ^m k k )g. The optmalty of ths transformaton s n the sense of mnmzng Equaton 3. The transformaton T s a 14

15 vector representng a rgd 3-D transformaton (rotaton + orentaton) of the model from ts local coordnates to the vewer coordnates. T s estmated from the generated uncertan lnes usng the Kalman lter tools (K.F). The estmaton process s composed of an ncremental renement, for whch at each step k ; 1, there exsts an estmate ^T k;1 of the transformaton T and a covarance matrx k;1 whch represents the \qualty" of ths estmate: k;1 = Ef( ^T k;1 ; T)( ^T k;1 ; T) t g : Gven a new measurement (^m k k ), an assocate uncertan lne, f( ^M k ; k )g, s generated, and the current estmate s updated to be ^T k wth an assocated uncertanty k. The accuracy of the estmate ncreases, as addtonal measurements are fused,.e. k k;1 ( k;1 ; k s nonnegatve dente). The process termnates as soon as no addtonal measurements can be suppled. 7.1 The Kalman Flter for Parameter Estmaton The Kalman lter s a tool for parameter estmaton from gven measurements. In our case the parameter vector to be estmated s the transformaton vector T whch s composed of two components: The translaton component, expressed by thevector t t =(t x t y t z ) t : (5) The rotaton component, descrbed by the quaternon ~q (see Appendx A) [1]: ~q =(q q) =(q q 1 + q 2 j + q 3 k) : The rotaton quaternon should satsfy the normalty constrans: ~q~q = q 2 + kqk2 =1, where ~q s the conjugate of ~q. In practce we represent the rotaton component by the vector: s q q from whch the quaternon ~q can be reconstructed: q = 1 p 1+st s ~q =(q q s) : 15

16 estmated-parameter ( ^, Ω ) T k-1 k-1 tme delay Kalman-flter ( m ^ ( T ^ k, Γ k) k, Ω k) measurement box result h k (x, m T) k k, mathematcal relatonshp Fgure 5: The Kalman lter for statc-parameter estmaton The three nputs and the estmaton output. The vector s s a convenent representaton of the rotatonal component n addton to beng mnmal (havng 3 parameters) the rotaton equaton s lnear n s as wll be shown later. In order to avod sngulartes n the representaton when q =we always use two object-centered coordnate systems, smultaneously. Consderng these two components, the parameter vector to be estmated durng the lterng process s: T = s t! : The Kalman lter produces an estmate ^T of the transformaton vector, gven the uncertan lnes. At each step, the lter receves three nputs and supples a sngle output (see Fgure 5). The nputs are: 1. An a pror estmate of the evaluated parameter vector and the uncertanty assocated wth t. In our case, n the k th step, the a pror estmaton wll be the estmate evaluated at the prevous step ^T k;1 and ts assocated covarance k;1. The covarance 16

17 matrx n the ntal step wll be set to nnty snce no a pror knowledge about T s assumed and the choce of ^T does not aect the end result. 2. The current measurement and ts uncertanty, n our case ths measurement s the uncertan lne ( ^M k ; k ) generated from actual 2D measurement (^m k k ) where ^M k and ts assocated covarance matrx ; k secton. s calculated as elaborated n the prevous 3. A mathematcal relatonshp between the evaluated parameters and the measurements. Ths mathematcal relatonshp should be lnear n the evaluated parameters. In our case the relatonshp s: ~M k = ~q ~ X k ~q + ~t (6) where ~ M k ~ X k ~t are quaternons assocated wth the vectors M k X k t respectvely. Gven that ~q~q =1,multplyng Equaton (6) by ~q yelds: ~M k ~q = ~q ~ X k + ~t~q : Isolatng the vector component of ths quaternon equaton and dvdng by q we obtan the matrx equaton: h k (X k M k T) <M k + X k> s +(M k ; X k) ; (I 3 ; <s>)t = (7) where s q q as prevously dened, I 3 s the 3 3 dentty matrx and <> denotes the matrx form of a cross product,.e: <v>= 1 C A <v> m = ; <m> v = v m : ;v z v y v z ;v x ;v y v x The equaton h k (X k M k T) = s not lnear as requred n the K.F., therefore, weuse the extended Kalman f lter (E.K.F.) [13, 16] whch s a generalzaton of the Kalman lter to non-lnear systems where transton from step k ; 1 to step k s performed usng a lnear approxmaton of h k by takng the rst order Taylor expanson around ( ^X k ^M k ^T k;1 ). The lnearzaton of the measurement equatons for our case are gven n Appendx B. 17

18 The output of the K.F fuser s an updated estmaton of the evaluated parameters and ts assocated uncertanty n our case ^T k and k respectvely. The K.F. fuser s of the form: ^T k = f( ^T k;1 k;1 ^X k k ^M k ; k h k ) : Thus, at each stage k, there s no need of retanng any of the prevously consdered measurements. Only the current estmate ^T k;1 and ts assocated uncertanty k;1 need be retaned. The K.F. updatng equatons are gven n Appendx C. The K.F. updatng equatons yeld an unbased estmate of T whch s optmal n the lnear mnmal varance crteron [1],.e. ^T k C = kx =1 mnmzes ^h (T)W ;1 ^ht (T) (8) where ^h (T) =h ( ^X ^M T) andw s the covarance matrx of ^h (Appendx D). The value C n Equaton 8 s n fact the same value as n Equaton 3 whch we amed to mnmze. Thus, the desred soluton as formulated n Sectons 3 and 4 s the same soluton as obtaned from the K.F. updatng equatons. Snce the measurement equatons are lnear approxmatons of non-lnear equatons the ntal soluton obtaned by the K.F. wll not necessarly be the correct soluton. Ths case wll happen when the lnearzaton s around a pont that s not close enough to the correct soluton. In order to reduce the nuence of the lnearzaton local teratve K:F: [13, page 349] s appled. In the teratons the constrants are relnearzed around the updated soluton obtaned by the K.F. and another cycle of K.F. s performed usng the new verson of the lnearzed constrants. Note, that the general K.F. deals wth a parameter vector that s changng wth tme, whereas n our case the estmated transformaton, T, s statc and does not change between measurements. 8 From Predcted Objects to Recognton After the optmal transformaton ^T n s estmated from the n measurements, the second stage n the recognton task s to nd the \smlarty" between the predcted object and the 18

19 canddate model. Ths \smlarty" s calculated wth the 3D dstance, however t has to be consder wth the uncertanty of the model features and the measurements. Fortunately, there s no need to reconstruct the predcted object n order to calculate ts smlarty to the model: Gven a transformaton T, the M.D. between the predcted object and the transformed model s gven by C of Equaton 8 where the summaton s over the n measurements. Thus, the smlarty between the predcted object and a canddate model s gven by thevalue C calculated for the nal transformaton ^T n. If a reconstructon of the predcted object s requred, the calculaton of ts feature locatons s smple: We transform each model-pont ^X by the optmal transformaton ^T n and nd apont U such that the sum of the M.D. to the uncertan lne ( ^M ; ) and the M.D. to the transformed model pont T ( ^X ) s mnmal. The mathematcal formulaton of ths reconstructon s elaborated n Appendx E. Note that f the th measurement s a perfect measurement ( s zero) then U wll be on the lne lne( ^m ). 9 Results We tested our method by smulatng a model as a collecton of ponts. The ponts of the model were chosen by random samplng n the cube [::1] 3. The model ponts were transformed by a transformaton T composed of a rotaton s and a translaton t. The model ponts and the measurements were contamnated by whte Gaussan nose. All model ponts were contamnated by 3D nose havng dentcal uncertanty (and dagonal covarance matrx) and all the measurements were contamnated wth 2D nose whch was added to the mage plane and havng dentcal uncertanty as well. Graphs 6-8 show the convergence of the estmates of the rotaton ^s and the translaton ^t as a functon of the number of measurements. The vertcal ordnate represents the squared devaton of the estmate from the real value,.e: t error = k^t ; tk 2 left graphs and s error = k^s ; sk 2 rght graphs : The results shown n the graphs were averaged over 1 processes of 1 randomly generated objects. In the graphs two cases are shown: the convergence of the estmate when 2D dstance metrc s used as the mnmzaton crteron, and the convergence when the 3D dstance s used. The varance of the nose was not gven to the algorthms and t was assumed that each measured pont s contamnated by the same amount. 19

20 Graphs 6-8 show three typcal cases: Graph 6 shows a case where the nose n the modelponts s domnant (the s.t.d of the model nose was proportonal to 8% of the total sze of the body where the mage nose was proportonal to 1%), Graph 7 shows a case where the nose n the mage s domnant (the s.t.d of the mage nose was proportonal to 5% of the total sze of the body where the model nose was proportonal to 1%), and Grpah 8 shows a case where the nose n the model-ponts s domnant (the same as n Graph 6) but the dstance of the body from the mage plane s qute large (1 tmes the focal length) relatve to the case shown n graphs 6 (where the dstance s about 1 tmes the focal length). It s demonstrated that the 3D dstance metrc s advantageous over the 2D dstance metrc when the model nose s domnant and the object s located qute close to the mage plane. In the cases where the mage nose s domnant or when the object s far away from the mage plane the mprovement of the estmate usng the 3D dstance metrc s neglgble. These results are reasonable snce the algorthms assume the same amount of nose for each measurement. When the mage nose s domnant ths assumpton s correct and the 2D metrc gves the same results as the 3D metrc. If the model nose s domnant, ts projecton onto the mage plane s not dentcal for each measurement and therefore 3D metrc s essental. In the case where the object s far away from the mage plane, the projecton of the model nose onto the mage plane s almost dentcal for all the measurements and therefore the the 2D metrc s agan useful. 2

21 D Eucldean metrc 2D Eucldean metrc.4 3D Eucldean metrc 2D Eucldean metrc R.M.S.E of t 6 4 R.M.S.E of s no. of ponts no. of ponts Fgure 6: A comparson between a 2D dstance metrc based algorthm and a 3D dstance metrc based algorthm. Left graph: convergence of the devaton of the translaton estmate ^t. Rght graph: convergence of the devaton of the rotaton estmate ^s. In ths case the s.t.d of the model nose was proportonal to 8% of the total sze of the body where the mage nose was proportonal to 1% D Eucldean metrc 2D Eucldean metrc.4 3D Eucldean metrc 2D Eucldean metrc R.M.S.E of t 6 4 R.M.S.E of s no. of ponts no. of ponts Fgure 7: A comparson between a 2D dstance metrc based algorthm and a 3D dstance metrc based algorthm. Left graph: convergence of the devaton of the translaton estmate ^t. Rght graph: convergence of the devaton of the rotaton estmate ^s. In ths case the s.t.d of the model nose was proportonal to 1% of the total sze of the body where the mage nose was proportonal to 5%. 21

22 D Eucldean metrc 2D Eucldean metrc.4 3D Eucldean metrc 2D Eucldean metrc R.M.S.E of t 6 4 R.M.S.E of s no. of ponts no. of ponts Fgure 8: A comparson between a 2D dstance metrc based algorthm and a 3D dstance metrc based algorthm. Left graph: convergence of the devaton of the translaton estmate ^t. Rght graph: convergence of the devaton of the rotaton estmate ^s. In ths case the s.t.d of the model nose and the mage nose were the same as n graphs 6, but the dstance of the body from the mage plane was about 1 tmes the focal length. 22

23 1 Concluson Ths paper presents a new scheme whch denes a 3D dstance metrc between model features and 2D measurements obtaned followng a projecton. Ths 3D dstance s dened as the mnmum 3D dstorton the model must undergo n order to t some permssable reconstructon of the measurements. It was demonstrated that the proposed 3D dstance metrc s usefull when the model nose s domnant and when the object s close to the mage plane. Acknowledgements I wsh to thank Ronen Basr for a frutful dscusson whch led me to ths paper. Appendx: A Rotaton Quaternon A quaternon ~q s composed of two parts - the scalar part q and the vector part q: ~q =(q q) =(q q 1 + q 2 j + q 3 k) : If v,v are vectors n < 3 such that v = Rv, when R s a rotaton matrx, then the correspondng expresson n quaternon form s: ~v = ~q~v~q : The quaternons ~v ~v correspond to the vectors v,v respectvely as follows: ~v =( v) ~v =( v ) and ~q s the conjugate of ~q ~q =(q ;q) : ~q represents a rotaton of the vector v by angle around a unt vector ^n where: q = cos( 2 ) q =sn( 2 )^n 23

24 so that k~qk 2 = ~q~q = q 2 + kqk 2 =1 : B Lnearzaton of the Measurement Equatons for the E.K.F At the k th step there s a non-lnear measurement equaton: h k (X k M k T) = as wrtten n Equaton 7. Snce the K.F. deals only wth lnear processes we use the lnear approxmaton of h k by takng the rst order Taylor expanson around ( ^X k ^M k ^T k;1 ): h k (X k M k T) = h k ( ^X k ^M k ^T k;1 k (X X k ; ^X k (M k ; M ^M k k k k T (T ; ^T k;1 ) (9) Equaton 9 can be rewrtten as a lnear equaton: where z k = H k T + k (1) z k = <^s k;1 > ^t k;1 + ^X k ; ^M k H k = [< ^M k + ^X k ; ^t k;1 > (<^s k;1 > ;I 3 )] k = [I 3 ; <^s k;1 >](M k ; ^M k ) ; [I 3 + <^s k;1 >](X k ; ^X k) : z k represents the new \measurement", H k s the matrx denotng a lnear connecton between the \measurement" and the actual transformaton T. Both z k and H k can be derved from ^M k, ^X k, ^T k;1. The term k - depcts the nose n the \measurement" z k and satses: Ef k g = varf k g = [I 3 ; <^s k;1 >] k [I 3 ; <^s k;1 >] t +[I 3 + <^s k;1 >] k [I 3 + <^s k;1 >] t = W k : Notce that accordng to the K.F. denton t s assumed that there s no correlaton between the derent measurement nose (covf j g =8 6= j). Ths assumpton s not always vald. When there s correlaton between several measurements, we may consder these measurements as a sngle measurement by groupng the measurement values nto a sngle vector and by combnng ther correspondng equatons nto a sngle vector equaton. 24

25 C The Kalman Flter Equatons for Statc Systems Assume the measurement equatons are as wrtten n Equaton 1. The recursve K.F. updatng equatons for tme step k are: state estmate update : ^T k = ^T k;1 + K k (z k ; H k ^T k;1 ) state covarance update : k = k;1 ; K k H k k;1 Kalman gan matrx : K k = k;1 Hk(H t k k;1 H t k + W k ) ;1 : D Covarance Matrx of ^h The value of h (Equaton 7) can be lnearly approxmated, n the th step, by takng the rst order Taylor expanson around ( ^X ^M ): h = ^h (X X ; ^X (M ; M ^M ) where the dervatves are taken at ( ^X ^M ) and ^h clear that ^h E s a zero mean random process wth the covarance: W = Ef^h ^ht g =(@h ) X ) t ); X ) t : M M Reconstructon of Predcted Object = h ( ^X ^M T). From the above ts Let ^T = ^s^t! be the nal transformaton estmated after fusng all the avalable measurements. From ^s we can buld an assocated rotaton matrx ^R [17] and an expresson for the transformed model: ^X = ^R ^X + ^t = ^R ^Rt Gven the transformed model-pont (^X ) and the assocated measured pont (^M ; ) (uncertan lne) the predcted object s obtan by fusng these two ponts: U = ^X + K( ^M ; ^X ) 25

26 where K = ( +; ) ;1 Ths equaton can be obtan from the K.F. updatng equatons where ^X state vector and ^M as a measurement. s regarded as the References [1] B.D.O. Anderson and J.B. Moore. Optmal Flterng. Prentce-Hall, Inc., N.J., [2] K.S. Arun, T.S. Huang, and S.D. Blosten. Least squares ttng of two 3D pont sets. IEEE Trans. Pattern Analyss and Machne Intellgence, 9(5):698{7, Sept [3] R. Basr and D. Wenshal. Dstance metrc between 3D models and 2D mages. In The 9th Israel Symposum on Artcal Intellgence and Computer Vson, pages 337{356, Dec [4] T. Fan, G. Medon, and R. Nevata. Recognzng 3D objects usng surface descrptons. IEEE Trans. Pattern Analyss and Machne Intellgence, 11(11):114{1157, Nov [5] O.D. Faugeras and M. Hebert. The representaton, recognton, and postonng of 3D shapes from range data. In A. Rosenfeld, edtor, Technques for 3D Machne Percepton, pages 13{51. Elsever Scence, [6] M.A. Fschler and R.C. Bolles. Random sample consensus: A paradgm for model ttng wth applcatons to mage analyss and automated cartography. Communcatons of the ACM, 24(6):381{395, June [7] W.E.L. Grmson. On the vercaton of hypotheszed matches n model-based recognton. In European Conference on Computer Vson, pages 489{498, 199. [8] W.E.L. Grmson and T. Lozano-Perez. Model-based recognton and localzaton from sparse range or tactle data. Internatonal Journal of Robotcs Research, 3(3):3{35, [9] Y. Hel-Or and M. Werman. Absolute orentaton from uncertan data: A uned approach. In Conference on Computer Vson and Pattern Recognton, pages 77{82, [1] B.K.P. Horn. Robot Vson. MIT Press,

27 [11] B.K.P Horn. Closed-form soluton of absolute orentaton usng unt quaternon. J. Opt. Soc. Am., 4(4):629{642, [12] D.P. Huttenlocher and S. Ullman. Object recognton usng alngment. In Internatonal Conf. on Computer Vson, pages 12{111, [13] A.H. Jazwnsk. Stochastc Process and Flterng Theory. Academc Press, 197. [14] D.G. Lowe. Three-dmensonal object recognton from sngle two-dmensonal mages. Artcal Intellgence, 31:355{395, [15] D.G. Lowe. Fttng parametrzed 3-D models to mages. IEEE Trans. Pattern Analyss and Machne Intellgence, 13:441{45, May [16] P.S. Maybeck. Stochastc Models, Estmaton, and Control, volume 1. Academc Press, [17] B. Sabata and J.K. Aggarwal. Estmaton of moton from a par of range mages: A revew. Computer Vson, Graphcs and Image Processng, 54(3):39{324, Nov

SLAM Summer School 2006 Practical 2: SLAM using Monocular Vision

SLAM Summer School 2006 Practical 2: SLAM using Monocular Vision SLAM Summer School 2006 Practcal 2: SLAM usng Monocular Vson Javer Cvera, Unversty of Zaragoza Andrew J. Davson, Imperal College London J.M.M Montel, Unversty of Zaragoza. josemar@unzar.es, jcvera@unzar.es,

More information

A Binarization Algorithm specialized on Document Images and Photos

A Binarization Algorithm specialized on Document Images and Photos A Bnarzaton Algorthm specalzed on Document mages and Photos Ergna Kavalleratou Dept. of nformaton and Communcaton Systems Engneerng Unversty of the Aegean kavalleratou@aegean.gr Abstract n ths paper, a

More information

R s s f. m y s. SPH3UW Unit 7.3 Spherical Concave Mirrors Page 1 of 12. Notes

R s s f. m y s. SPH3UW Unit 7.3 Spherical Concave Mirrors Page 1 of 12. Notes SPH3UW Unt 7.3 Sphercal Concave Mrrors Page 1 of 1 Notes Physcs Tool box Concave Mrror If the reflectng surface takes place on the nner surface of the sphercal shape so that the centre of the mrror bulges

More information

Fitting & Matching. Lecture 4 Prof. Bregler. Slides from: S. Lazebnik, S. Seitz, M. Pollefeys, A. Effros.

Fitting & Matching. Lecture 4 Prof. Bregler. Slides from: S. Lazebnik, S. Seitz, M. Pollefeys, A. Effros. Fttng & Matchng Lecture 4 Prof. Bregler Sldes from: S. Lazebnk, S. Setz, M. Pollefeys, A. Effros. How do we buld panorama? We need to match (algn) mages Matchng wth Features Detect feature ponts n both

More information

Range images. Range image registration. Examples of sampling patterns. Range images and range surfaces

Range images. Range image registration. Examples of sampling patterns. Range images and range surfaces Range mages For many structured lght scanners, the range data forms a hghly regular pattern known as a range mage. he samplng pattern s determned by the specfc scanner. Range mage regstraton 1 Examples

More information

3D vector computer graphics

3D vector computer graphics 3D vector computer graphcs Paolo Varagnolo: freelance engneer Padova Aprl 2016 Prvate Practce ----------------------------------- 1. Introducton Vector 3D model representaton n computer graphcs requres

More information

Analysis of Continuous Beams in General

Analysis of Continuous Beams in General Analyss of Contnuous Beams n General Contnuous beams consdered here are prsmatc, rgdly connected to each beam segment and supported at varous ponts along the beam. onts are selected at ponts of support,

More information

CS 534: Computer Vision Model Fitting

CS 534: Computer Vision Model Fitting CS 534: Computer Vson Model Fttng Sprng 004 Ahmed Elgammal Dept of Computer Scence CS 534 Model Fttng - 1 Outlnes Model fttng s mportant Least-squares fttng Maxmum lkelhood estmaton MAP estmaton Robust

More information

Subspace clustering. Clustering. Fundamental to all clustering techniques is the choice of distance measure between data points;

Subspace clustering. Clustering. Fundamental to all clustering techniques is the choice of distance measure between data points; Subspace clusterng Clusterng Fundamental to all clusterng technques s the choce of dstance measure between data ponts; D q ( ) ( ) 2 x x = x x, j k = 1 k jk Squared Eucldean dstance Assumpton: All features

More information

Skew Angle Estimation and Correction of Hand Written, Textual and Large areas of Non-Textual Document Images: A Novel Approach

Skew Angle Estimation and Correction of Hand Written, Textual and Large areas of Non-Textual Document Images: A Novel Approach Angle Estmaton and Correcton of Hand Wrtten, Textual and Large areas of Non-Textual Document Images: A Novel Approach D.R.Ramesh Babu Pyush M Kumat Mahesh D Dhannawat PES Insttute of Technology Research

More information

Computer Animation and Visualisation. Lecture 4. Rigging / Skinning

Computer Animation and Visualisation. Lecture 4. Rigging / Skinning Computer Anmaton and Vsualsaton Lecture 4. Rggng / Sknnng Taku Komura Overvew Sknnng / Rggng Background knowledge Lnear Blendng How to decde weghts? Example-based Method Anatomcal models Sknnng Assume

More information

Parallelism for Nested Loops with Non-uniform and Flow Dependences

Parallelism for Nested Loops with Non-uniform and Flow Dependences Parallelsm for Nested Loops wth Non-unform and Flow Dependences Sam-Jn Jeong Dept. of Informaton & Communcaton Engneerng, Cheonan Unversty, 5, Anseo-dong, Cheonan, Chungnam, 330-80, Korea. seong@cheonan.ac.kr

More information

Structure from Motion

Structure from Motion Structure from Moton Structure from Moton For now, statc scene and movng camera Equvalentl, rgdl movng scene and statc camera Lmtng case of stereo wth man cameras Lmtng case of multvew camera calbraton

More information

Support Vector Machines

Support Vector Machines /9/207 MIST.6060 Busness Intellgence and Data Mnng What are Support Vector Machnes? Support Vector Machnes Support Vector Machnes (SVMs) are supervsed learnng technques that analyze data and recognze patterns.

More information

Robust Computation and Parametrization of Multiple View. Relations. Oxford University, OX1 3PJ. Gaussian).

Robust Computation and Parametrization of Multiple View. Relations. Oxford University, OX1 3PJ. Gaussian). Robust Computaton and Parametrzaton of Multple Vew Relatons Phl Torr and Andrew Zsserman Robotcs Research Group, Department of Engneerng Scence Oxford Unversty, OX1 3PJ. Abstract A new method s presented

More information

Improvement of Spatial Resolution Using BlockMatching Based Motion Estimation and Frame. Integration

Improvement of Spatial Resolution Using BlockMatching Based Motion Estimation and Frame. Integration Improvement of Spatal Resoluton Usng BlockMatchng Based Moton Estmaton and Frame Integraton Danya Suga and Takayuk Hamamoto Graduate School of Engneerng, Tokyo Unversty of Scence, 6-3-1, Nuku, Katsuska-ku,

More information

A Fast Visual Tracking Algorithm Based on Circle Pixels Matching

A Fast Visual Tracking Algorithm Based on Circle Pixels Matching A Fast Vsual Trackng Algorthm Based on Crcle Pxels Matchng Zhqang Hou hou_zhq@sohu.com Chongzhao Han czhan@mal.xjtu.edu.cn Ln Zheng Abstract: A fast vsual trackng algorthm based on crcle pxels matchng

More information

Delayed Features Initialization for Inverse Depth Monocular SLAM

Delayed Features Initialization for Inverse Depth Monocular SLAM Delayed Features Intalzaton for Inverse Depth Monocular SLAM Rodrgo Mungua and Anton Grau Department of Automatc Control, Techncal Unversty of Catalona, UPC c/ Pau Gargallo, 5 E-0808 Barcelona, Span, {rodrgo.mungua;anton.grau}@upc.edu

More information

Detection of an Object by using Principal Component Analysis

Detection of an Object by using Principal Component Analysis Detecton of an Object by usng Prncpal Component Analyss 1. G. Nagaven, 2. Dr. T. Sreenvasulu Reddy 1. M.Tech, Department of EEE, SVUCE, Trupath, Inda. 2. Assoc. Professor, Department of ECE, SVUCE, Trupath,

More information

Outline. Type of Machine Learning. Examples of Application. Unsupervised Learning

Outline. Type of Machine Learning. Examples of Application. Unsupervised Learning Outlne Artfcal Intellgence and ts applcatons Lecture 8 Unsupervsed Learnng Professor Danel Yeung danyeung@eee.org Dr. Patrck Chan patrckchan@eee.org South Chna Unversty of Technology, Chna Introducton

More information

Reducing Frame Rate for Object Tracking

Reducing Frame Rate for Object Tracking Reducng Frame Rate for Object Trackng Pavel Korshunov 1 and We Tsang Oo 2 1 Natonal Unversty of Sngapore, Sngapore 11977, pavelkor@comp.nus.edu.sg 2 Natonal Unversty of Sngapore, Sngapore 11977, oowt@comp.nus.edu.sg

More information

An efficient method to build panoramic image mosaics

An efficient method to build panoramic image mosaics An effcent method to buld panoramc mage mosacs Pattern Recognton Letters vol. 4 003 Dae-Hyun Km Yong-In Yoon Jong-Soo Cho School of Electrcal Engneerng and Computer Scence Kyungpook Natonal Unv. Abstract

More information

Active Contours/Snakes

Active Contours/Snakes Actve Contours/Snakes Erkut Erdem Acknowledgement: The sldes are adapted from the sldes prepared by K. Grauman of Unversty of Texas at Austn Fttng: Edges vs. boundares Edges useful sgnal to ndcate occludng

More information

Feature Reduction and Selection

Feature Reduction and Selection Feature Reducton and Selecton Dr. Shuang LIANG School of Software Engneerng TongJ Unversty Fall, 2012 Today s Topcs Introducton Problems of Dmensonalty Feature Reducton Statstc methods Prncpal Components

More information

12/2/2009. Announcements. Parametric / Non-parametric. Case-Based Reasoning. Nearest-Neighbor on Images. Nearest-Neighbor Classification

12/2/2009. Announcements. Parametric / Non-parametric. Case-Based Reasoning. Nearest-Neighbor on Images. Nearest-Neighbor Classification Introducton to Artfcal Intellgence V22.0472-001 Fall 2009 Lecture 24: Nearest-Neghbors & Support Vector Machnes Rob Fergus Dept of Computer Scence, Courant Insttute, NYU Sldes from Danel Yeung, John DeNero

More information

A Robust Method for Estimating the Fundamental Matrix

A Robust Method for Estimating the Fundamental Matrix Proc. VIIth Dgtal Image Computng: Technques and Applcatons, Sun C., Talbot H., Ourseln S. and Adraansen T. (Eds.), 0- Dec. 003, Sydney A Robust Method for Estmatng the Fundamental Matrx C.L. Feng and Y.S.

More information

A Comparison and Evaluation of Three Different Pose Estimation Algorithms In Detecting Low Texture Manufactured Objects

A Comparison and Evaluation of Three Different Pose Estimation Algorithms In Detecting Low Texture Manufactured Objects Clemson Unversty TgerPrnts All Theses Theses 12-2011 A Comparson and Evaluaton of Three Dfferent Pose Estmaton Algorthms In Detectng Low Texture Manufactured Objects Robert Krener Clemson Unversty, rkrene@clemson.edu

More information

MULTISPECTRAL IMAGES CLASSIFICATION BASED ON KLT AND ATR AUTOMATIC TARGET RECOGNITION

MULTISPECTRAL IMAGES CLASSIFICATION BASED ON KLT AND ATR AUTOMATIC TARGET RECOGNITION MULTISPECTRAL IMAGES CLASSIFICATION BASED ON KLT AND ATR AUTOMATIC TARGET RECOGNITION Paulo Quntlano 1 & Antono Santa-Rosa 1 Federal Polce Department, Brasla, Brazl. E-mals: quntlano.pqs@dpf.gov.br and

More information

Mathematics 256 a course in differential equations for engineering students

Mathematics 256 a course in differential equations for engineering students Mathematcs 56 a course n dfferental equatons for engneerng students Chapter 5. More effcent methods of numercal soluton Euler s method s qute neffcent. Because the error s essentally proportonal to the

More information

Smoothing Spline ANOVA for variable screening

Smoothing Spline ANOVA for variable screening Smoothng Splne ANOVA for varable screenng a useful tool for metamodels tranng and mult-objectve optmzaton L. Rcco, E. Rgon, A. Turco Outlne RSM Introducton Possble couplng Test case MOO MOO wth Game Theory

More information

Cluster Analysis of Electrical Behavior

Cluster Analysis of Electrical Behavior Journal of Computer and Communcatons, 205, 3, 88-93 Publshed Onlne May 205 n ScRes. http://www.scrp.org/ournal/cc http://dx.do.org/0.4236/cc.205.350 Cluster Analyss of Electrcal Behavor Ln Lu Ln Lu, School

More information

Hermite Splines in Lie Groups as Products of Geodesics

Hermite Splines in Lie Groups as Products of Geodesics Hermte Splnes n Le Groups as Products of Geodescs Ethan Eade Updated May 28, 2017 1 Introducton 1.1 Goal Ths document defnes a curve n the Le group G parametrzed by tme and by structural parameters n the

More information

Fitting: Deformable contours April 26 th, 2018

Fitting: Deformable contours April 26 th, 2018 4/6/08 Fttng: Deformable contours Aprl 6 th, 08 Yong Jae Lee UC Davs Recap so far: Groupng and Fttng Goal: move from array of pxel values (or flter outputs) to a collecton of regons, objects, and shapes.

More information

Lobachevsky State University of Nizhni Novgorod. Polyhedron. Quick Start Guide

Lobachevsky State University of Nizhni Novgorod. Polyhedron. Quick Start Guide Lobachevsky State Unversty of Nzhn Novgorod Polyhedron Quck Start Gude Nzhn Novgorod 2016 Contents Specfcaton of Polyhedron software... 3 Theoretcal background... 4 1. Interface of Polyhedron... 6 1.1.

More information

Accounting for the Use of Different Length Scale Factors in x, y and z Directions

Accounting for the Use of Different Length Scale Factors in x, y and z Directions 1 Accountng for the Use of Dfferent Length Scale Factors n x, y and z Drectons Taha Soch (taha.soch@kcl.ac.uk) Imagng Scences & Bomedcal Engneerng, Kng s College London, The Rayne Insttute, St Thomas Hosptal,

More information

Multi-stable Perception. Necker Cube

Multi-stable Perception. Necker Cube Mult-stable Percepton Necker Cube Spnnng dancer lluson, Nobuuk Kaahara Fttng and Algnment Computer Vson Szelsk 6.1 James Has Acknowledgment: Man sldes from Derek Hoem, Lana Lazebnk, and Grauman&Lebe 2008

More information

Face Recognition University at Buffalo CSE666 Lecture Slides Resources:

Face Recognition University at Buffalo CSE666 Lecture Slides Resources: Face Recognton Unversty at Buffalo CSE666 Lecture Sldes Resources: http://www.face-rec.org/algorthms/ Overvew of face recognton algorthms Correlaton - Pxel based correspondence between two face mages Structural

More information

Exterior Orientation using Coplanar Parallel Lines

Exterior Orientation using Coplanar Parallel Lines Exteror Orentaton usng Coplanar Parallel Lnes Frank A. van den Heuvel Department of Geodetc Engneerng Delft Unversty of Technology Thsseweg 11, 69 JA Delft, The Netherlands Emal: F.A.vandenHeuvel@geo.tudelft.nl

More information

Edge Detection in Noisy Images Using the Support Vector Machines

Edge Detection in Noisy Images Using the Support Vector Machines Edge Detecton n Nosy Images Usng the Support Vector Machnes Hlaro Gómez-Moreno, Saturnno Maldonado-Bascón, Francsco López-Ferreras Sgnal Theory and Communcatons Department. Unversty of Alcalá Crta. Madrd-Barcelona

More information

Learning the Kernel Parameters in Kernel Minimum Distance Classifier

Learning the Kernel Parameters in Kernel Minimum Distance Classifier Learnng the Kernel Parameters n Kernel Mnmum Dstance Classfer Daoqang Zhang 1,, Songcan Chen and Zh-Hua Zhou 1* 1 Natonal Laboratory for Novel Software Technology Nanjng Unversty, Nanjng 193, Chna Department

More information

Unsupervised Learning and Clustering

Unsupervised Learning and Clustering Unsupervsed Learnng and Clusterng Why consder unlabeled samples?. Collectng and labelng large set of samples s costly Gettng recorded speech s free, labelng s tme consumng 2. Classfer could be desgned

More information

New Extensions of the 3-Simplex for Exterior Orientation

New Extensions of the 3-Simplex for Exterior Orientation New Extensons of the 3-Smplex for Exteror Orentaton John M. Stenbs Tyrone L. Vncent Wllam A. Hoff Colorado School of Mnes jstenbs@gmal.com tvncent@mnes.edu whoff@mnes.edu Abstract Object pose may be determned

More information

Unsupervised Learning

Unsupervised Learning Pattern Recognton Lecture 8 Outlne Introducton Unsupervsed Learnng Parametrc VS Non-Parametrc Approach Mxture of Denstes Maxmum-Lkelhood Estmates Clusterng Prof. Danel Yeung School of Computer Scence and

More information

Fusion of Data from Head-Mounted and Fixed Sensors 1

Fusion of Data from Head-Mounted and Fixed Sensors 1 Frst Internatonal Workshop on Augmented Realty, Nov., 998, San Francsco. Fuson of Data from Head-ounted and Fxed Sensors Abstract Wllam A. Hoff Engneerng Dvson, Colorado School of nes Golden, Colorado

More information

LECTURE : MANIFOLD LEARNING

LECTURE : MANIFOLD LEARNING LECTURE : MANIFOLD LEARNING Rta Osadchy Some sldes are due to L.Saul, V. C. Raykar, N. Verma Topcs PCA MDS IsoMap LLE EgenMaps Done! Dmensonalty Reducton Data representaton Inputs are real-valued vectors

More information

Large Motion Estimation for Omnidirectional Vision

Large Motion Estimation for Omnidirectional Vision Large Moton Estmaton for Omndrectonal Vson Jong Weon Lee, Suya You, and Ulrch Neumann Computer Scence Department Integrated Meda Systems Center Unversty of Southern Calforna Los Angeles, CA 98978, USA

More information

y and the total sum of

y and the total sum of Lnear regresson Testng for non-lnearty In analytcal chemstry, lnear regresson s commonly used n the constructon of calbraton functons requred for analytcal technques such as gas chromatography, atomc absorpton

More information

Content Based Image Retrieval Using 2-D Discrete Wavelet with Texture Feature with Different Classifiers

Content Based Image Retrieval Using 2-D Discrete Wavelet with Texture Feature with Different Classifiers IOSR Journal of Electroncs and Communcaton Engneerng (IOSR-JECE) e-issn: 78-834,p- ISSN: 78-8735.Volume 9, Issue, Ver. IV (Mar - Apr. 04), PP 0-07 Content Based Image Retreval Usng -D Dscrete Wavelet wth

More information

An Optimal Algorithm for Prufer Codes *

An Optimal Algorithm for Prufer Codes * J. Software Engneerng & Applcatons, 2009, 2: 111-115 do:10.4236/jsea.2009.22016 Publshed Onlne July 2009 (www.scrp.org/journal/jsea) An Optmal Algorthm for Prufer Codes * Xaodong Wang 1, 2, Le Wang 3,

More information

Robust Recovery of Camera Rotation from Three Frames. B. Rousso S. Avidan A. Shashua y S. Peleg z. The Hebrew University of Jerusalem

Robust Recovery of Camera Rotation from Three Frames. B. Rousso S. Avidan A. Shashua y S. Peleg z. The Hebrew University of Jerusalem Robust Recovery of Camera Rotaton from Three Frames B. Rousso S. Avdan A. Shashua y S. Peleg z Insttute of Computer Scence The Hebrew Unversty of Jerusalem 994 Jerusalem, Israel e-mal : roussocs.huj.ac.l

More information

Problem Set 3 Solutions

Problem Set 3 Solutions Introducton to Algorthms October 4, 2002 Massachusetts Insttute of Technology 6046J/18410J Professors Erk Demane and Shaf Goldwasser Handout 14 Problem Set 3 Solutons (Exercses were not to be turned n,

More information

FEATURE EXTRACTION. Dr. K.Vijayarekha. Associate Dean School of Electrical and Electronics Engineering SASTRA University, Thanjavur

FEATURE EXTRACTION. Dr. K.Vijayarekha. Associate Dean School of Electrical and Electronics Engineering SASTRA University, Thanjavur FEATURE EXTRACTION Dr. K.Vjayarekha Assocate Dean School of Electrcal and Electroncs Engneerng SASTRA Unversty, Thanjavur613 41 Jont Intatve of IITs and IISc Funded by MHRD Page 1 of 8 Table of Contents

More information

Simultaneous Object Pose and Velocity Computation Using a Single View from a Rolling Shutter Camera

Simultaneous Object Pose and Velocity Computation Using a Single View from a Rolling Shutter Camera Smultaneous Object Pose and Velocty Computaton Usng a Sngle Vew from a Rollng Shutter Camera Omar At-Ader, Ncolas Andreff, Jean Marc Lavest, and Phlppe Martnet Unversté Blase Pascal Clermont Ferrand, LASMEA

More information

Ecient Computation of the Most Probable Motion from Fuzzy. Moshe Ben-Ezra Shmuel Peleg Michael Werman. The Hebrew University of Jerusalem

Ecient Computation of the Most Probable Motion from Fuzzy. Moshe Ben-Ezra Shmuel Peleg Michael Werman. The Hebrew University of Jerusalem Ecent Computaton of the Most Probable Moton from Fuzzy Correspondences Moshe Ben-Ezra Shmuel Peleg Mchael Werman Insttute of Computer Scence The Hebrew Unversty of Jerusalem 91904 Jerusalem, Israel Emal:

More information

Tsinghua University at TAC 2009: Summarizing Multi-documents by Information Distance

Tsinghua University at TAC 2009: Summarizing Multi-documents by Information Distance Tsnghua Unversty at TAC 2009: Summarzng Mult-documents by Informaton Dstance Chong Long, Mnle Huang, Xaoyan Zhu State Key Laboratory of Intellgent Technology and Systems, Tsnghua Natonal Laboratory for

More information

Lecture 5: Multilayer Perceptrons

Lecture 5: Multilayer Perceptrons Lecture 5: Multlayer Perceptrons Roger Grosse 1 Introducton So far, we ve only talked about lnear models: lnear regresson and lnear bnary classfers. We noted that there are functons that can t be represented

More information

Lecture 4: Principal components

Lecture 4: Principal components /3/6 Lecture 4: Prncpal components 3..6 Multvarate lnear regresson MLR s optmal for the estmaton data...but poor for handlng collnear data Covarance matrx s not nvertble (large condton number) Robustness

More information

Parallel matrix-vector multiplication

Parallel matrix-vector multiplication Appendx A Parallel matrx-vector multplcaton The reduced transton matrx of the three-dmensonal cage model for gel electrophoress, descrbed n secton 3.2, becomes excessvely large for polymer lengths more

More information

Line-based Camera Movement Estimation by Using Parallel Lines in Omnidirectional Video

Line-based Camera Movement Estimation by Using Parallel Lines in Omnidirectional Video 01 IEEE Internatonal Conference on Robotcs and Automaton RverCentre, Sant Paul, Mnnesota, USA May 14-18, 01 Lne-based Camera Movement Estmaton by Usng Parallel Lnes n Omndrectonal Vdeo Ryosuke kawansh,

More information

An Application of the Dulmage-Mendelsohn Decomposition to Sparse Null Space Bases of Full Row Rank Matrices

An Application of the Dulmage-Mendelsohn Decomposition to Sparse Null Space Bases of Full Row Rank Matrices Internatonal Mathematcal Forum, Vol 7, 2012, no 52, 2549-2554 An Applcaton of the Dulmage-Mendelsohn Decomposton to Sparse Null Space Bases of Full Row Rank Matrces Mostafa Khorramzadeh Department of Mathematcal

More information

Image Alignment CSC 767

Image Alignment CSC 767 Image Algnment CSC 767 Image algnment Image from http://graphcs.cs.cmu.edu/courses/15-463/2010_fall/ Image algnment: Applcatons Panorama sttchng Image algnment: Applcatons Recognton of object nstances

More information

Amnon Shashua Shai Avidan Michael Werman. The Hebrew University, objects.

Amnon Shashua Shai Avidan Michael Werman. The Hebrew University,   objects. Trajectory Trangulaton over Conc Sectons Amnon Shashua Sha Avdan Mchael Werman Insttute of Computer Scence, The Hebrew Unversty, Jerusalem 91904, Israel e-mal: fshashua,avdan,wermang@cs.huj.ac.l Abstract

More information

TN348: Openlab Module - Colocalization

TN348: Openlab Module - Colocalization TN348: Openlab Module - Colocalzaton Topc The Colocalzaton module provdes the faclty to vsualze and quantfy colocalzaton between pars of mages. The Colocalzaton wndow contans a prevew of the two mages

More information

A mathematical programming approach to the analysis, design and scheduling of offshore oilfields

A mathematical programming approach to the analysis, design and scheduling of offshore oilfields 17 th European Symposum on Computer Aded Process Engneerng ESCAPE17 V. Plesu and P.S. Agach (Edtors) 2007 Elsever B.V. All rghts reserved. 1 A mathematcal programmng approach to the analyss, desgn and

More information

IMPROVING AND EXTENDING THE INFORMATION ON PRINCIPAL COMPONENT ANALYSIS FOR LOCAL NEIGHBORHOODS IN 3D POINT CLOUDS

IMPROVING AND EXTENDING THE INFORMATION ON PRINCIPAL COMPONENT ANALYSIS FOR LOCAL NEIGHBORHOODS IN 3D POINT CLOUDS IMPROVING AND EXTENDING THE INFORMATION ON PRINCIPAL COMPONENT ANALYSIS FOR LOCAL NEIGHBORHOODS IN 3D POINT CLOUDS Davd Belton Cooperatve Research Centre for Spatal Informaton (CRC-SI) The Insttute for

More information

Kinematics of pantograph masts

Kinematics of pantograph masts Abstract Spacecraft Mechansms Group, ISRO Satellte Centre, Arport Road, Bangalore 560 07, Emal:bpn@sac.ernet.n Flght Dynamcs Dvson, ISRO Satellte Centre, Arport Road, Bangalore 560 07 Emal:pandyan@sac.ernet.n

More information

MOTION PANORAMA CONSTRUCTION FROM STREAMING VIDEO FOR POWER- CONSTRAINED MOBILE MULTIMEDIA ENVIRONMENTS XUNYU PAN

MOTION PANORAMA CONSTRUCTION FROM STREAMING VIDEO FOR POWER- CONSTRAINED MOBILE MULTIMEDIA ENVIRONMENTS XUNYU PAN MOTION PANORAMA CONSTRUCTION FROM STREAMING VIDEO FOR POWER- CONSTRAINED MOBILE MULTIMEDIA ENVIRONMENTS by XUNYU PAN (Under the Drecton of Suchendra M. Bhandarkar) ABSTRACT In modern tmes, more and more

More information

Covariance-Based Registration

Covariance-Based Registration Covarance-Based Regstraton Charles V. Stewart Dept. of Computer Scence Rensselaer Poly. Inst. Troy, New York 12180 stewart@cs.rp.edu June 24, 2002 1 Introducton The regstraton problem n computer vson s

More information

CORRELATION ICP ALGORITHM FOR POSE ESTIMATION BASED ON LOCAL AND GLOBAL FEATURES

CORRELATION ICP ALGORITHM FOR POSE ESTIMATION BASED ON LOCAL AND GLOBAL FEATURES CORRELATION ICP ALGORITHM FOR POSE ESTIMATION BASED ON LOCAL AND GLOBAL FEATURES Marco A. Chavarra, Gerald Sommer Cogntve Systems Group. Chrstan-Albrechts-Unversty of Kel, D-2498 Kel, Germany {mc,gs}@ks.nformatk.un-kel.de

More information

Corner-Based Image Alignment using Pyramid Structure with Gradient Vector Similarity

Corner-Based Image Alignment using Pyramid Structure with Gradient Vector Similarity Journal of Sgnal and Informaton Processng, 013, 4, 114-119 do:10.436/jsp.013.43b00 Publshed Onlne August 013 (http://www.scrp.org/journal/jsp) Corner-Based Image Algnment usng Pyramd Structure wth Gradent

More information

S1 Note. Basis functions.

S1 Note. Basis functions. S1 Note. Bass functons. Contents Types of bass functons...1 The Fourer bass...2 B-splne bass...3 Power and type I error rates wth dfferent numbers of bass functons...4 Table S1. Smulaton results of type

More information

Type-2 Fuzzy Non-uniform Rational B-spline Model with Type-2 Fuzzy Data

Type-2 Fuzzy Non-uniform Rational B-spline Model with Type-2 Fuzzy Data Malaysan Journal of Mathematcal Scences 11(S) Aprl : 35 46 (2017) Specal Issue: The 2nd Internatonal Conference and Workshop on Mathematcal Analyss (ICWOMA 2016) MALAYSIAN JOURNAL OF MATHEMATICAL SCIENCES

More information

Assignment # 2. Farrukh Jabeen Algorithms 510 Assignment #2 Due Date: June 15, 2009.

Assignment # 2. Farrukh Jabeen Algorithms 510 Assignment #2 Due Date: June 15, 2009. Farrukh Jabeen Algorthms 51 Assgnment #2 Due Date: June 15, 29. Assgnment # 2 Chapter 3 Dscrete Fourer Transforms Implement the FFT for the DFT. Descrbed n sectons 3.1 and 3.2. Delverables: 1. Concse descrpton

More information

IMAGE MATCHING WITH SIFT FEATURES A PROBABILISTIC APPROACH

IMAGE MATCHING WITH SIFT FEATURES A PROBABILISTIC APPROACH IMAGE MATCHING WITH SIFT FEATURES A PROBABILISTIC APPROACH Jyot Joglekar a, *, Shrsh S. Gedam b a CSRE, IIT Bombay, Doctoral Student, Mumba, Inda jyotj@tb.ac.n b Centre of Studes n Resources Engneerng,

More information

Vanishing Hull. Jinhui Hu, Suya You, Ulrich Neumann University of Southern California {jinhuihu,suyay,

Vanishing Hull. Jinhui Hu, Suya You, Ulrich Neumann University of Southern California {jinhuihu,suyay, Vanshng Hull Jnhu Hu Suya You Ulrch Neumann Unversty of Southern Calforna {jnhuhusuyay uneumann}@graphcs.usc.edu Abstract Vanshng ponts are valuable n many vson tasks such as orentaton estmaton pose recovery

More information

Calibrating a single camera. Odilon Redon, Cyclops, 1914

Calibrating a single camera. Odilon Redon, Cyclops, 1914 Calbratng a sngle camera Odlon Redon, Cclops, 94 Our goal: Recover o 3D structure Recover o structure rom one mage s nherentl ambguous??? Sngle-vew ambgut Sngle-vew ambgut Rashad Alakbarov shadow sculptures

More information

Programming in Fortran 90 : 2017/2018

Programming in Fortran 90 : 2017/2018 Programmng n Fortran 90 : 2017/2018 Programmng n Fortran 90 : 2017/2018 Exercse 1 : Evaluaton of functon dependng on nput Wrte a program who evaluate the functon f (x,y) for any two user specfed values

More information

Fuzzy Filtering Algorithms for Image Processing: Performance Evaluation of Various Approaches

Fuzzy Filtering Algorithms for Image Processing: Performance Evaluation of Various Approaches Proceedngs of the Internatonal Conference on Cognton and Recognton Fuzzy Flterng Algorthms for Image Processng: Performance Evaluaton of Varous Approaches Rajoo Pandey and Umesh Ghanekar Department of

More information

What are the camera parameters? Where are the light sources? What is the mapping from radiance to pixel color? Want to solve for 3D geometry

What are the camera parameters? Where are the light sources? What is the mapping from radiance to pixel color? Want to solve for 3D geometry Today: Calbraton What are the camera parameters? Where are the lght sources? What s the mappng from radance to pel color? Why Calbrate? Want to solve for D geometry Alternatve approach Solve for D shape

More information

The Research of Ellipse Parameter Fitting Algorithm of Ultrasonic Imaging Logging in the Casing Hole

The Research of Ellipse Parameter Fitting Algorithm of Ultrasonic Imaging Logging in the Casing Hole Appled Mathematcs, 04, 5, 37-3 Publshed Onlne May 04 n ScRes. http://www.scrp.org/journal/am http://dx.do.org/0.436/am.04.584 The Research of Ellpse Parameter Fttng Algorthm of Ultrasonc Imagng Loggng

More information

LEAST SQUARES. RANSAC. HOUGH TRANSFORM.

LEAST SQUARES. RANSAC. HOUGH TRANSFORM. LEAS SQUARES. RANSAC. HOUGH RANSFORM. he sldes are from several sources through James Has (Brown); Srnvasa Narasmhan (CMU); Slvo Savarese (U. of Mchgan); Bll Freeman and Antono orralba (MI), ncludng ther

More information

Feature Extraction and Registration An Overview

Feature Extraction and Registration An Overview Feature Extracton and Regstraton An Overvew S. Seeger, X. Laboureux Char of Optcs, Unversty of Erlangen-Nuremberg, Staudstrasse 7/B2, 91058 Erlangen, Germany Emal: sns@undne.physk.un-erlangen.de, xl@undne.physk.un-erlangen.de

More information

Positive Semi-definite Programming Localization in Wireless Sensor Networks

Positive Semi-definite Programming Localization in Wireless Sensor Networks Postve Sem-defnte Programmng Localzaton n Wreless Sensor etworks Shengdong Xe 1,, Jn Wang, Aqun Hu 1, Yunl Gu, Jang Xu, 1 School of Informaton Scence and Engneerng, Southeast Unversty, 10096, anjng Computer

More information

6.1 2D and 3D feature-based alignment 275. similarity. Euclidean

6.1 2D and 3D feature-based alignment 275. similarity. Euclidean 6.1 2D and 3D feature-based algnment 275 y translaton smlarty projectve Eucldean affne x Fgure 6.2 Basc set of 2D planar transformatons Once we have extracted features from mages, the next stage n many

More information

NUMERICAL SOLVING OPTIMAL CONTROL PROBLEMS BY THE METHOD OF VARIATIONS

NUMERICAL SOLVING OPTIMAL CONTROL PROBLEMS BY THE METHOD OF VARIATIONS ARPN Journal of Engneerng and Appled Scences 006-017 Asan Research Publshng Network (ARPN). All rghts reserved. NUMERICAL SOLVING OPTIMAL CONTROL PROBLEMS BY THE METHOD OF VARIATIONS Igor Grgoryev, Svetlana

More information

Angle-Independent 3D Reconstruction. Ji Zhang Mireille Boutin Daniel Aliaga

Angle-Independent 3D Reconstruction. Ji Zhang Mireille Boutin Daniel Aliaga Angle-Independent 3D Reconstructon J Zhang Mrelle Boutn Danel Alaga Goal: Structure from Moton To reconstruct the 3D geometry of a scene from a set of pctures (e.g. a move of the scene pont reconstructon

More information

MOTION BLUR ESTIMATION AT CORNERS

MOTION BLUR ESTIMATION AT CORNERS Gacomo Boracch and Vncenzo Caglot Dpartmento d Elettronca e Informazone, Poltecnco d Mlano, Va Ponzo, 34/5-20133 MILANO boracch@elet.polm.t, caglot@elet.polm.t Keywords: Abstract: Pont Spread Functon Parameter

More information

Inverse-Polar Ray Projection for Recovering Projective Transformations

Inverse-Polar Ray Projection for Recovering Projective Transformations nverse-polar Ray Projecton for Recoverng Projectve Transformatons Yun Zhang The Center for Advanced Computer Studes Unversty of Lousana at Lafayette yxz646@lousana.edu Henry Chu The Center for Advanced

More information

SIMULTANEOUS REGISTRATION OF MULTIPLE VIEWS OF A 3D OBJECT

SIMULTANEOUS REGISTRATION OF MULTIPLE VIEWS OF A 3D OBJECT SIMULTANEOUS REGISTRATION OF MULTIPLE VIEWS OF A 3D OBJECT Helmut Pottmann a, Stefan Leopoldseder a, Mchael Hofer a a Insttute of Geometry, Venna Unversty of Technology, Wedner Hauptstr. 8 10, A 1040 Wen,

More information

Registration of Expressions Data using a 3D Morphable Model

Registration of Expressions Data using a 3D Morphable Model Regstraton of Expressons Data usng a 3D Morphable Model Curzo Basso DISI, Unverstà d Genova, Italy Emal: curzo.basso@ds.unge.t Thomas Vetter Departement Informatk, Unverstät Basel, Swtzerland Emal: thomas.vetter@unbas.ch

More information

Multi-view 3D Position Estimation of Sports Players

Multi-view 3D Position Estimation of Sports Players Mult-vew 3D Poston Estmaton of Sports Players Robbe Vos and Wlle Brnk Appled Mathematcs Department of Mathematcal Scences Unversty of Stellenbosch, South Afrca Emal: vosrobbe@gmal.com Abstract The problem

More information

RELATIVE ORIENTATION ESTIMATION OF VIDEO STREAMS FROM A SINGLE PAN-TILT-ZOOM CAMERA. Commission I, WG I/5

RELATIVE ORIENTATION ESTIMATION OF VIDEO STREAMS FROM A SINGLE PAN-TILT-ZOOM CAMERA. Commission I, WG I/5 RELATIVE ORIENTATION ESTIMATION OF VIDEO STREAMS FROM A SINGLE PAN-TILT-ZOOM CAMERA Taeyoon Lee a, *, Taeung Km a, Gunho Sohn b, James Elder a a Department of Geonformatc Engneerng, Inha Unersty, 253 Yonghyun-dong,

More information

Recognizing Faces. Outline

Recognizing Faces. Outline Recognzng Faces Drk Colbry Outlne Introducton and Motvaton Defnng a feature vector Prncpal Component Analyss Lnear Dscrmnate Analyss !"" #$""% http://www.nfotech.oulu.f/annual/2004 + &'()*) '+)* 2 ! &

More information

REFRACTION. a. To study the refraction of light from plane surfaces. b. To determine the index of refraction for Acrylic and Water.

REFRACTION. a. To study the refraction of light from plane surfaces. b. To determine the index of refraction for Acrylic and Water. Purpose Theory REFRACTION a. To study the refracton of lght from plane surfaces. b. To determne the ndex of refracton for Acrylc and Water. When a ray of lght passes from one medum nto another one of dfferent

More information

An Accurate Evaluation of Integrals in Convex and Non convex Polygonal Domain by Twelve Node Quadrilateral Finite Element Method

An Accurate Evaluation of Integrals in Convex and Non convex Polygonal Domain by Twelve Node Quadrilateral Finite Element Method Internatonal Journal of Computatonal and Appled Mathematcs. ISSN 89-4966 Volume, Number (07), pp. 33-4 Research Inda Publcatons http://www.rpublcaton.com An Accurate Evaluaton of Integrals n Convex and

More information

CMPS 10 Introduction to Computer Science Lecture Notes

CMPS 10 Introduction to Computer Science Lecture Notes CPS 0 Introducton to Computer Scence Lecture Notes Chapter : Algorthm Desgn How should we present algorthms? Natural languages lke Englsh, Spansh, or French whch are rch n nterpretaton and meanng are not

More information

Homography Estimation with L Norm Minimization Method

Homography Estimation with L Norm Minimization Method Homography Estmaton wth L orm Mnmzaton Method Hyunjung LEE Yongdue SEO and Rchard Hartley : Department o Meda echnology Sogang Unversty, Seoul, Korea, whtetob@sogang.ac.r : Department o Meda echnology

More information

An Iterative Solution Approach to Process Plant Layout using Mixed Integer Optimisation

An Iterative Solution Approach to Process Plant Layout using Mixed Integer Optimisation 17 th European Symposum on Computer Aded Process Engneerng ESCAPE17 V. Plesu and P.S. Agach (Edtors) 2007 Elsever B.V. All rghts reserved. 1 An Iteratve Soluton Approach to Process Plant Layout usng Mxed

More information

Simulation: Solving Dynamic Models ABE 5646 Week 11 Chapter 2, Spring 2010

Simulation: Solving Dynamic Models ABE 5646 Week 11 Chapter 2, Spring 2010 Smulaton: Solvng Dynamc Models ABE 5646 Week Chapter 2, Sprng 200 Week Descrpton Readng Materal Mar 5- Mar 9 Evaluatng [Crop] Models Comparng a model wth data - Graphcal, errors - Measures of agreement

More information

A Fast Content-Based Multimedia Retrieval Technique Using Compressed Data

A Fast Content-Based Multimedia Retrieval Technique Using Compressed Data A Fast Content-Based Multmeda Retreval Technque Usng Compressed Data Borko Furht and Pornvt Saksobhavvat NSF Multmeda Laboratory Florda Atlantc Unversty, Boca Raton, Florda 3343 ABSTRACT In ths paper,

More information