Non-Parametric Structure-Based Calibration of Radially Symmetric Cameras

Size: px
Start display at page:

Download "Non-Parametric Structure-Based Calibration of Radially Symmetric Cameras"

Transcription

1 Non-Parametrc Structure-Based Calbraton of Radally Symmetrc Cameras Federco Camposeco, Torsten Sattler, Marc Pollefeys Department of Computer Scence, ETH Zürch, Swtzerland {federco.camposeco, torsten.sattler, Abstract We propose a novel two-step method for estmatng the ntrnsc and extrnsc calbraton of any radally symmetrc camera, ncludng non-central systems. The frst step conssts of estmatng the camera pose, gven a Structure from Moton (SfM) model, up to the translaton along the optcal axs. As a second step, we obtan the calbraton by fndng the translaton of the camera center usng an orderng constrant. The method makes use of the 1D radal camera model, whch allows us to effectvely handle any radally symmetrc camera, ncludng non-central ones. Usng ths orderng constrant, we show that the we are able to calbrate several dfferent (central and non-central) Wde Feld of Vew (WFOV) cameras, ncludng fsheye, hypercatadoptrc and sphercal catadoptrc cameras, as well as pnhole cameras, usng a sngle mage or jontly solvng for several vews. 1. Introducton Recently, cameras wth a wde feld of vew (WFOV) such as fsheye and omndrectonal cameras are startng to become more and more popular. Due to ther ablty to observe a large porton of the scene, usng WFOV cameras s advantageous for 3D computer vson tasks such as the precse camera trackng that s done as part of vsual navgaton for robots and autonomous vehcles. At the same tme, acton cameras wth WFOV such as the GoPro Hero are wdely used. Smlarly, camera mounts for moble devces that enable them to take panoramc mages, e.g., usng a catadoptrc lens, are becomng more frequent. As a result, more and more WFOV mages are becomng avalable on photo sharng webstes such as Flckr and Pcasa. In the case of Structure-from-Moton (SfM) from photo communty collectons, these photos could be partcularly helpful to strengthen the overall reconstructon as they provde constrants to many other cameras. However, they are typcally dscarded n practce due to the challenge of automatcally calbratng these cameras. In ths paper, we present a novel method to automatcally calbrate WFOV cameras from 2D-3D matches establshed between features extracted n ther mages and 3D ponts n a SfM reconstructon usng mage-based localzaton methods. Gven a partal reconstructon obtaned from regular mages, our method can thus be used to calbrate and then nsert WFOV nto a SfM model to strongly lnk together dfferent parts of the scene whch are all vsble n these photos and thus mprove the qualty of the reconstructon. Our method s based on the 1D radal camera model [15 17], whch can be used to descrbe any type of camera wth radal dstorton, ncludng pnhole, fsheye, and non-central cameras such as catadoptrc lenses, as long as the center of dstorton s known. We combne ths model wth a nonparametrc ntrnsc calbraton to obtan an extremely powerful calbraton method that s capable of calbratng a wde range of camera types. Our method conssts of two steps. In the frst step, the extrnsc calbraton s computed from the 2D-3D matches up to the poston of the camera center (or camera centers n the case of non-central cameras) along the optcal axs. Ths can be done effcently usng a lnear 7-pont solver nsde a RANSAC loop [5]. Gven the partal extrnsc calbraton, we employ a novel orderng constrant on the openng angle of the vewng rays correspondng to the 2D features to estmate the remanng extrnsc parameter n the case of a central camera. Snce we are consderng radally symmetrc cameras, fxng the camera center drectly provdes the ntrnsc calbraton as t provdes a mappng from mage postons to vewng rays. A slghtly generalzed verson of ths constrant enables us to obtan a more accurate calbraton from multple photos taken wth the same camera. The constrant can be further generalzed to also handle non-central cameras, both when only a sngle or multple photos are avalable. In addton to ts generalty, our novel orderng constrant enables us to formulate the second part of the calbraton process as a convex optmzaton program. We expermentally demonstrate the accuracy of our calbraton ppelne for a wde range of cameras. Addtonally, we make the source code of our method avalable [1]. The remander of the paper s structured as follows. Sec. 2 dscusses related work. Sec. 3 revews the 1D radal camera model. Sec. 4 ntroduces our novel orderng constrant and derves our calbraton method for the case of central cameras. The extenson to non-central cameras 2192

2 s then provded n Sec. 5. Fnally, Sec. 6 expermentally evaluates our approach on both synthetc and real data. 2. Related Work Recently, there has been some nterest n non-parametrc calbraton of radally symmetrc cameras. For nstance, n [8, 11] a non-teratve, non-parametrc method for calbraton of fsheye cameras s proposed. Whle they clam that they can do wthout, the method s only tested usng a calbraton pattern snce ther auto-calbraton needs several correspondences and s very senstve to nose. Smlar to ths, and much more related to our work, n [15 17] Thrthala and Pollefeys developed the 1D Radal Camera model. They propose a mult-focal tensor able to auto-calbrate any radally symmetrc set of cameras (ncludng non-central), and also produce a non-parametrc calbraton. In ths work we use ths same model, however, n contrast to ther approach, we develop a more general and robust geometrc orderng constrant to calbrate wth. Several other methods make use of ether a specfc scene structure (enough straght lnes) or use calbraton objects to compute ther calbraton, e.g. [7, 13]. In [14] an approach for self-calbraton of radally symmetrc cameras s presented. They develop a plumb-lne (usng the fact that straght lnes n space must project nto straght lnes n the mage) and plane-based methods. In [6], they rely on the observaton of at least three lnes to compute the parameters of a para-catadoptrc system. In contrast to these, we reman flexble by enforcng no requrements on the scene or a calbraton object. For our method we frst estmate the (partal) extrnscs (.e. the pose) w.r.t. a SfM model. There have been numerous advances geared toward pose estmaton n the absence of calbraton. For example, Kukelova et al. propose a 5-pont pose estmator wth unknown radal dstorton and focal length [10]. They employ an dea smlar to the 1D Radal Camera Model n order to effcently obtan a pose and 3-parameter calbraton. Also, mnmal 4-pont solvers for ths same case are presented n [3, 9] n whch they too make use of a parametrc model to obtan a soluton. In contrast to these methods, we do not restrct the camera to be able to be represented by a specfc mode. Furthermore, we extend our method to seamlessly aggregate data from several vews of the same camera to ncrease the accuracy of the calbraton and to better handle non-central systems. 3. The 1D Radal Camera Model In order to calbrate any type of radally symmetrc camera, ths paper bulds on the 1D radal camera model. For a more n-depth analyss the reader s referred do [17], however we brefly revew t n ths secton for completeness. LetC d be the center of dstorton for a camera exhbtng radal dstorton. Let x u denote the undstorted projecton of a 3D pont X onto the camera s mage. As llustrated Fgure 1: Radal 1D Camera. Image plane (left) and top vew (rght) of the projecton of pontx. n Fg. 1, applyng radal dstorton maps x u to a pont on the radal lnel = x u C d through the center of dstorton and the undstorted mage coordnates. Smlarly,x u les on the lnel = x d C d defned by the dstorted measurement x d. Instead of explctly modelng the radal dstorton, the 1D radal camera model defnes a projecton up to radal dstorton. Ths s expressed as a mappng P 3 P 1 that assocates each 3D pont to a lneλl = P r X. The projecton matrxp r R 2 4 relates to the frst two rows of the camera pose (R t) by [ ][ ] 0 1 R1 t P r = x, (1) 1 0 R 2 t y where R s the -th row of the rotaton matrx R. Notce that the 1D radal camera model, other than unt aspect rato, makes no assumpton on the nternal calbraton of the camera. In fact, t descrbes both central and non-central cameras as long as there s a sngle center of dstorton, ncludng pnhole, fsheye, and catadoptrc cameras. As n [17], we assume that C d s known, enablng us to center the mage around C d. For most cameras, the center of the mage s a reasonable approxmaton forc d. Alternatvely, t can be estmated usng the vsble rm of the catadoptrc mrror or the edge of the fsheye lens (c.f. Fg. 8). 4. Calbratng Central Radally Symmetrc Cameras Gven a 3D model of the scene, our goal s to estmate both the extrnsc and ntrnsc calbraton from 2D-3D correspondences(x,x ) between postons n an mage taken wth a radally symmetrc camera and the model. Snce the projecton matrx P r does not depend on the ntrnsc calbraton, we use a two-stage approach. In the frst stage, we use RANSAC [5] to estmate the extrnsc calbraton up to an unknown translaton along the optcal axs. The nlers to the pose are then used to non-parametrcally estmate the ntrnsc calbraton. Sec. 4.1 detals the computaton of the partal extrnsc calbraton. In Sec. 4.2, we then derve a novel orderng constrant that allows us to compute the ntrnsc calbraton by solvng a convex optmzaton problem. Sec. 4.3 shows that the same constrant can be used to calbrate a camera from multple mages. We show n Sec

3 how to extend our approach to handle non-central radally symmetrc cameras Partal Extrnsc Calbraton Let x d = (x d,y d ) be the poston of a dstorted measurement n a coordnate system centered at the center of dstorton. The radal lne of the -th correspondence (x d,x ) can then be expressed as [ ] [ ] y l = d /x d l = = P 1 1 r X. (2) By multplyng l by ts perpendcular vector (1, l ), we obtan P r1 X l ( P r2 X ) = 0, (3) (a) Sngle mage central case. (b) Mult-mage central case. where P rn represents the n-th row of the matrx P r. Thus, each 2D-3D correspondence gves us one constrant. Snce P r s only defned up to scale, t can be estmated lnearly from seven matches by rearrangng (3). Once we have an estmate for P r, we can recover the full rotaton matrx R by explotng the fact that rotaton matrces are orthonormal matrces wth determnant one (c.f. (1)). Gven a set of 2D-3D correspondences, we estmate P r by usng the 7-pont solver nsde a RANSAC loop. In order to dstngush between nlers and outlers, we measure the subtended angle between the predcted and the observed radal lnesˆl = P r X and l. A match s consdered to be an nler f the angle s below a gven threshold σ (set to 1 n our experments). Notce that P r has only fve degrees of freedom n total: Three degrees of freedom for the rotaton and two degrees of freedom for the partal translaton t x, t y. Thus, the lnear 7-pont solver s non-mnmal. If a mnmal solver s requred due to a hgh outler rato, the 5-pont approach from [10] can be used, whch requres solvng a fourth degree polynomal n a sngle varable Non Parametrc Intrnsc Calbraton The ntrnsc calbraton of a camera defnes a mappng r(x) from mage coordnates to vewng rays. In the case of radal symmetry, the angle θ between the ray r(x) and the optcal axs for all postonsxwth the same dstance to the center of dstorton,.e., x 2 = r, s constant. Consequently, the pont X projectng to x has to le on a cone along the optcal axs wth openng angleθ (c.f. Fg. 1). For two ponts x r1, x r2 wth rad r 1 < r 2, we have θ r1 < θ r2. In the followng, we derve a geometrc constrant from ths observaton from whch we explctly compute the mappng from rad to openng angles. Gven P r (c.f. Sec. 4.1), the transformaton of the 3D ponts from the global nto the local coordnate system of the camera s defned up to a translaton along the optcal axs. Usng R, t = (t x,t y,0), we obtan an ntermedate (c) Sngle mage non-central. (d) Mult-mage non-central. Fgure 2: Orderng constrants for dfferent systems. The abscssa for each fgure, labeled z, are algned wth the optcal axs of the camera. For Fgures a, b and c,r d > rj d. coordnate system n whch the unknown translaton corresponds to the poston of the camera center c on the optcal axs. We notce that fxng c defnes the openng angle θ for a gven 3D pontrx+t n the ntermedate coordnate system. Thus, fxng c fully defnes the ntrnsc calbraton of the camera. A geometrc orderng constrant on the camera center. We express each pont(x,y,ẑ ) = (x,y,z c) as (ϕ,ρ,ẑ ) n a cylndrcal coordnate system (c.f. Fg. 2). Snce we consder radally symmetrc cameras, we can drop the angle ϕ of the pont around the optcal axs from the notaton and only consder the dstance of the 3D pont to the optcal axs ρ and ts depth z. Consder two 3D ponts p = (ρ,ẑ ), p j = (ρ j,ẑ j ) n the ntermedate frame (c.f. Fg. 2a), correspondng to rad rd and rj d of the dstorted mage measurements x d, x j d. Wthout loss of generalty, letρ ρ j and let I j = (ẑ j ρ ẑ ρ j )/(ρ ρ j ) (4) be the ntersecton of the 2D lne contanng the pont par wth the optcal axs z. In the case where rd = rj d, I j corresponds to the camera center c. Unfortunately, t s rather unlkely to fnd two features wth exactly the same radus. In [15], the authors propose to ft a lne through 3D ponts correspondng to smlar rad to obtan a camera center per radus. In contrast, we use an orderng constrant to drectly obtan acas explaned below. Wthout loss of generalty, assume thatrd > rj d and thus θ > θ j. In the case thatρ > ρ j, t follows thatc < I j (c.f. Fg. 2a). Smlarly, ρ < ρ j yelds the constrant c > I j. 2194

4 Thus, for each pont par we get a one-sded constrant that restrcts the value ofcto le ether to the left or to the rght of I j. For each constrant we then buld a cost functon whch penalzes a gven c that volates a one-sded constrant by usng a pecewse cost functon. Forrd > rj d and ρ > ρ j { Ej l 0 c < I j (c;p,p j ) = f (I j c) otherwse, (5) whch penalzes c f t s to the rght of I j (c.f. Fg 2a). Here, f s a functon dependng on the dstance between the ntersecton pont and the center c. For the opposte confguraton, ether r d < rj d or ρ < ρ j, we may buld a smlar cost functon whch penalzes c to be to the left of I j. Calbraton through convex optmzaton. Usng (5) we can then take the sum over all the cost functons E(c) = Ej l + Ej r, (6) {,j} L {,j} R where L = {{,j} ρ > ρ j and rd > rj d } and R = {{,j} ρ > ρ j and rd < rj d }. If f s chosen to be a convex functon, E(c) wll be convex. We can then obtan the camera center, and thus the ntrnsc calbraton, by optmzng (6) usng, e.g., Gradent-Descent. We choose f to be an L1 norm to be robust to outlers, whle allowng E to reman convex. Furthermore, we propose a very smple algorthm for computng (6) whenf s a lnear functon. Snce the slope of E(c) changes only wherever there s an ntersecton, we may effcently compute t n two passes. We start by sortng the ntersectons, such thati k < I k+1. On the frst pass, from left to rght, we deal only wth the ntersectons that constran c to be to ther left (shown n green n Fg. 3a) and teratvely compute the cost for each ntersecton. Startng wthe(i 0 ) l = 0 we can express the cost of thek th ntersecton as the cost of the prevous ntersecton plus the cost ncrease of the k volatng constrants fromi k 1 toi k. Sncef s a lnear functon, the latter costs only depend on the dstance between the current ntersecton and the last,.e. E(I k ) l = E(I k 1 ) l +kf (I k I k 1 ). (7) On the second pass we sweep n the opposte drecton takng nto account the ntersectons that constran c to be to ther rght. The cost of a gven c k s obtaned by checkng ts nearest left- and rght-constranng ntersectons and summng ther costse(c k ) = E r (I k )+E l (I k ). Selectng pont pars. GvenN ponts n an mage, t s mpractcal to exhaustvely take all pont pars snce the number of pars s N(N 1)/2. Instead, for a gven mage we want to only operate on a fxed number of pars. To do so, we to sort the pont pars by ther qualty,.e. pars whch yeld stable ntersectons close to c. For each par p j we (a) Fgure 3: In a we show the sngle mage lnear cost functon proposed. In the mult-mage case, each pont par defnes a 2D constrant, depcted n b. get r j = r d rj d and ρ j = ρ ρ j. Frst, we dscard pars wth ρ j less than a gven threshold, whch takes care of unstable ntersectons. Then we sort the pars usng r j n ascendng order and take only the frstn s (set to 120 n our experments) pars of the sorted lst. (b) 4.3. Jont Calbraton from Multple Images The approach presented n Sec. 4.2 essentally determnes an nterval n whch the camera center can le n. Usng more ponts adds more constrants on ths nterval, whch should lead to more accurate estmates. Synthetc experments have shown that approxmately 250 maged ponts are enough to obtan a calbraton that acheves less than 1 pxels of RMSE on the reprojected ponts (c.f. Fg. 5), whle usng less than 100 ponts leaves c very underconstraned and the resultng calbraton wll be unrelable 1. Naturally, addtonal ponts can be obtaned by usng multple mages for the calbraton. Thus, n ths secton we show that our geometrc orderng constrant can easly be extended to allow calbratng a camera from M > 1 mages. By expressng 3D ponts n the ρz-plane we can transform all cameras to a common frame of reference by fndng a one-dmensonal relatve translaton between them (c.f. Fg. 2b). Ths allows us to employ our one-sded constrant to fnd ths relatve translaton and a jont calbraton. Jont constrant for central cameras. Gven two cameras s() and s(j) we can express the ntersecton of any pont parp j between them as I s(),s(j) = ( z j c s(j) ) ρ ( z c s() ) ρj, (8) where s() ndcates to whch camera the pont corresponds. Notce that ths s almost the same as (4), however (8) provdes a constrant that now depends on two varables (c.f. Fg. 3b), makng the poston of one camera dependent on the other. The cost functon E : R M R can be also desgned as a pecewse functon. For the confguraton 1 Notce, that toolboxes as the one descrbed n [13] suggest usng 6 to 10 mages. Assumng a calbraton pattern wth 48 corners, such methods use up to 480 ponts. 2195

5 r d > rj d and ρ > ρ j we defne E l j ( ) {0 I cs(),c s(j) = s(),s(j) < 0 g ( ) c s(),c s(j) otherwse, (9) where g s a cost functon on the dstance from the gven center par to the ntersecton (c.f. Fg. 2b). As wth (5), we decde to use the L1 norm as a cost functon to reman robust to outlers. Calbraton through convex optmzaton. Smlarly to (6), we take the sum of all relevant pont pars and to get E, whch can be mnmzed usng a convex optmzaton method. Notce that the selecton crtera for pont pars descrbed n Sec. 4.1 apples here as well, snce we may aggregate mage rad from all cameras nto one sngle sorted lst to choose N s relevant pars. Fnally, we get a calbraton by translatng the Z coordnate of the ponts ( ) by the camera center that obtaned the observaton θ r d = arctan ( ρ /(z c s() ) ). 5. Calbratng Non-Central Cameras Snce we are only dealng wth radally symmetrc cameras, the centers of the camera can be expressed as a functon of the dstorted mage radus c = c(r d ). So, any pont p = (ρ,ẑ ) hasẑ = z c, wherec = c(r d ). Non-central constrant. Any pont par p j wll constran both centersc andc j (see Fg. 2c). GvenN mage ponts, we have N 1 constrants for each center we need to estmate. We treat each of the N camera centers as a dfferent vew of the scene (.e. s() = ) and apply the method descrbed n Sec However, n practce the N 1 constrants mght not lmt the locaton of a gven center enough, yeldng naccurate results for centers wth weak or too few constrants (e.g. for centers that correspond to rad near the edge or center of the mage, snce these are mostly same-sde constrants). To solve ths we propose to mpose an orderng constrant to the centers. We frst sort all the ponts p by ther rad such that r 1 d < rd < r+1 d, whch restrcts ther correspondng centers c ( r 1 d ) ( ) ( ) < c r d < c r +1 d. (10) Ths s sensble gven that all radally symmetrc non-central systems known to the authors follow ths orderng (e.g. sphercal catadoptrc, para-catadoptrc). Ths constrant can seamlessly be translated nto the one-sded constrants (c.f. Fg. 4). We defne E r k(c) = N ( E r k (c)+ek(c) l ), (11) k=0.e. the cost of the k-th pont aganst the rest. Then for a Fgure 4: Illustraton of the orderng constrant from smulated data. In red, each center s constraned to le above a certan value, and vce versa for blue. Notce that we may use any of the blue constrants lyng to the left of any gven pxel radus and vce versa. centerc the cost becomes E(c ) = N E r k(c)+ E l k(c), (12) k=0 k= n other words, we use the one-sded left constrants of the centers that should be larger thanc and the one-sded rght constrants of those centers that should be smaller than c (see Fg. 2c). Mnmzng (12) we get a set of centers whch can be used to get the fnal calbraton mappng θ ( r d ) = arctan(ρ /(z c )). Jont soluton for non-central cameras. To get a jont non-central calbraton we use a two step procedure. Frst, we treat each camera as a central system and solve for ther jont calbraton, whch provdes us wth an estmate of ther dsplacements d k (see Fg. 2d). We use ths to translate all the data ponts across dfferent vews to be on the same frame of reference,.e. to have a mutually consstent depth. Second, we solve for a sngle non-central system by treatng all the translated ponts as f they came from a sngle vew. Ths allows us to keep the number of ponts needed for a successful calbraton relatvely low (around 350 ponts per mage, c.f. Fg. 5) Refnement and Fnal Calbraton One of the prmary benefts of our method s that we provde a calbraton that does not rely on a gven parametrzaton, thus we can accommodate a very wde range of cameras; from planar to catadoptrc, central and non-central. However, we wsh to refne our obtaned soluton by removng vews and ponts based on ther reprojecton errors, and to do so we must fnd a way to use the obtaned mappng. For ths we opt to use a sldng medan [8] of the calbraton 2196

6 Fgure 5: RMSE error when varyng the number of matches used to obtan the calbraton (pxel σ set to 1.2). The red astersk ndcates that pror to that number of matches, the calbraton faled. (a) Equangular. (b) FOV. data obtaned 2. For all vews we compute correspondng reprojecton errors and remove those ponts whose errors rse beyond a certan threshold (set to 5 pxels n our experments). After ths, we get a fnal set of nlers and recompute the calbraton by repeatng the correspondng procedure. 6. Expermental Evaluaton To evaluate the proposed method we perform experments wth real and synthetc data. Snce one of the strengths of the method s that t can handle a very wde array of cameras, we make a pont of tryng as many cameras as possble (c.f. Fg. 8) Synthetc Data We frst carred out experments on synthetc data to evaluate the performance of our methods. We populate the scene wth 320 data ponts dstrbuted randomly. To smulate the central case, we project the data ponts nto the camera usng a pnhole model as well as two well-known fsheye models, the Feld of Vew (FOV) model [4] and the equangular model. For the non-central case, we chose a sphercal catadoptrc camera. To compute the reflectons we use [2]. Fg. 7 shows the calbraton output of the central as well as the non-central smulatons, Fg. 6 compares the accuracy of the generated calbratons aganst those obtaned usng the toolbox n [13], and Fg. 5 shows the error w.r.t the number of matches used. As t can be seen from Fg. 7, the results for the central systems perfectly match the ground truth. In Fg. 7c we show the beneft of relaxng the method to handle noncentral systems. At the begnnng of the curve both orange (central assumpton) and blue (non-central) scatter plots match. However, as the non-centralty of the sphercal model becomes more sgnfcant at hgher rad of the smulated mage, the devaton s more apparent. In Fg. 7d we show how the accuracy of c(r d ) s affected when we do not enforce the orderng constrant (10). We show the re- 2 However, havng a calbraton that s agnostc to the partcular optcs of the setup, one s free to use a more sophstcated method to approxmate the dstorton functon. (c) Sphercal Catadoptrc. Fgure 6: Comparson of the reprojecton error and ts standard devaton aganst [13]. To obtan the calbratons, our method used 320 ponts whle the method we compare aganst used 21 mages of a 48-pont synthetc calbraton pattern. Notce that for Fg. 6c the error s always lower usng our calbraton snce we explctly support non-central systems. sultng calbraton mappngs compared aganst the ground truth of the smulated data. Ths s of partcular mportance for the smulated sphercal catadoptrc system snce wth real data we don t have relable ground truth forc(r d ) Real Data In order to test the flexblty of the method, we tred several dfferent cameras and lenses (c.f. Fg. 8): A Nkon D300 coupled wth a fsheye lens, a 360One VR catadoptrc lens and a sphercal catadoptrc lens (usng a 3-nch steel ball). To assess the performance for the mentoned moble phone attachments, we run tests usng an Phone 4 wth a GoPano catadoptrc attachment. Results from other cameras and lenses are provded as supplemental materal [1]. Due to the hgh dstorton observed wth most of the lenses, we don t obtan a very hgh number of matches and thus we must use the mult-camera methods. To get more complete calbratons n the catadoptrc cases we need to ncrease the number of matches near the edges of the reflecton. To do so, we frst use an equangular calbraton: θ = k r d where we fnd k by havng θ = π/2 map to the largest rad n the mage. We use ths to warp the mage nto a cylndrcal map whch we use to get matches aganst the SfM model 3. Note that ths places no restrctons on the 3 The toolbox used for comparson had to be manually asssted to get the corners of the calbraton pattern for ths partcularly dffcult data. 2197

7 (a) Nkon D300 wth fsheye lens. (b) Phone 4 wth GoPano catadoptrc lens. (c) Sphercal reflecton wth Nkon D300. (d) 360One VR catadoptrc lens wth Nkon D300. Fgure 9: Real calbratons for central and non-central cases. All results shown are compared aganst the calbraton obtaned usng [13], shown n gray. As t can be seen from all four cameras, our soluton accurately match the one from ths stateof-the-art calbraton toolbox. Calbraton ponts are colored accordng to the mage ndex used to emphasze the number of mages used per camera. To hghlght the comparson wth the reference calbraton, we also plot the dstorton functon F (the plot on the rght for each case) as detaled n [13]. mages we are able to handle, snce the only assumpton, as before, s that the mages are radally symmetrc. The SfM model used conssts of a large-scale reconstructon, obtaned n an outdoor locaton, c.f. Fg. 10. We took several mages wth each camera type at the same locaton and obtaned putatve 3D-2D matches (around 120 for each mage). In order to maxmze the number of matches obtaned, we employed the method proposed n [12], modfed to return as many matches as possble. Because of the drastcally large dstorton, we observed nler ratos as low as 20%, and thus several mages were needed for each camera type (between 20 and 25) from whch we were left wth approxmately 500 ponts. In Fg. 9 we show the calbraton obtaned for a selecton of the tested systems. For each case, we obtaned calbratons that closely match the calbraton computed wth the toolbox n [13]. To emphasze the correctness of the obtaned calbraton, we compare the prevously dscussed mappng (θ(r d )) and the functon F. Ths functon s descrbed n [13] as the focal length as a functon of the mage radus, where a pont wth mage coordnates (u,v) can be expressed n the camera frame as (u,v,f(r d )). For the fsheye mages (c.f. Fg. 9a), we got a very hgh number of matches, snce the query mages resembled the most to the mages used to construct the SfM model. For ths partcular model, we have enough data to see that the calbraton near the center of the mage suffers more than the rest due to weak constrants. However, for the case of the non-central system (c.f. Fg. 9c) there s a larger msmatch throughout between our obtaned calbraton and that of [13] snce our method fully supports non-central systems. For the GoPano attachment, as shown n Fg. 9b, we have very few matches near the border and ths s reflected n the scattered data ponts at the end of the curve. Nevertheless, we are overall able to calbrate even such a low-qualty lens system. The reference calbraton method addtonally computes the refned centers of dstorton. However, the fact that our calbraton closely matches the reference calbraton shows that usng the center of the mage as the center of dstorton s a vald assumpton n practce. 7. Concluson In ths work we presented a novel, flexble, structurebased calbraton method for radally symmetrc cameras. Indeed, such subset of cameras encompasses most of the systems used nowadays, such as planar, fsheye, catadoptrc, WFOV, and so on. We are thus able to handle the cal- 2198

8 (a) Equangular θ(rd ). (b) FOV θ(rd ). (c) Sphercal θ(rd ). (d) Sphercal c(rd ). Fgure 7: Synthetc calbraton θ(rd ) for the central (a and b) and the non-central cameras c, where n orange we show how a central assumpton would not be as accurate. Fg. d shows the estmate of c(rd ) for the non-central case, notce here the effect of employng the orderng constrant (10). braton of several systems under a sngle framework whch would usually requre several dfferent calbraton methods. Furthermore, WFOV magery s becomng more ubqutous by products such as the GoPro and WFOV lens attachments for moble phones. Wth our method, we can make use of ths ncreasngly popular mage modalty to augment and strengthen SfM models produced from onlne photocollectons. Onlne WFOV mages can be thus calbrated and nserted as part of an exstng SfM model. Ths would be greatly benefcal for the qualty of the model snce these type of mages can strongly lnk several parts of the model whch were never vsble before from the same vew. The descrbed calbraton method makes use of the 1D Radal Camera [15] to decouple the estmaton of the extrnsc (up to translaton along the optcal axs) and ntrnsc calbraton of any radally symmetrc camera nto two separate steps. In partcular, the partal extrnscs are obtaned va a lnear 7-pont solver n conjuncton wth RANSAC, whle the computaton of the ntrnscs s carred out mnmzng an outler-robust convex cost functon for both the sngle and the mult-mage case. We compute the calbraton as a mappng from dstorted mage rad nto the angle of ts correspondng 3D ray w.r.t. the optcal axs of the camera. By optng for a nonparametrc calbraton we are able to mantan a very broad compatblty wth any camera that fts the 1D Radal model. Fgure 8: Sample of the data used for the experments for the same outdoor locaton. From top to bottom, left to rght; the 360One catadoptrc lens, the sphercal catadoptrc setup, the GoPano Phone attachment and the D300 wth a fsheye lens. Notce the wde range of dstortons as well as the low qualty n the case of the GoPano. Fgure 10: The SfM pont cloud used for the expermental evaluaton. The approach s valdated expermentally and usng real data and ts accuracy and robustness s assessed by comparng the obtaned calbraton mappngs aganst the calbraton from a state-of-the-art toolbox [13]. We make our source code avalable at [1]. Acknowledgements The research leadng to these results has receved fundng from Google s Project Tango. We would lke to thank Dr. Martn Oswald for hs valuable nput regardng the convex optmzaton. 2199

9 References [1] Project page. research/radally-symmetrc-cameras/. [2] A. Agrawal, Y. Taguch, and S. Ramalngam. Analytcal forward projecton for axal non-central doptrc and catadoptrc cameras. In ECCV, [3] M. Bujnak, Z. Kukelova, and T. Pajdla. New effcent soluton to the absolute pose problem for camera wth unknown focal length and radal dstorton. In ACCV [4] F. Devernay and O. Faugeras. Straght lnes have to be straght. Machne vson and applcatons, 13(1):14 24, [5] M. A. Fschler and R. C. Bolles. Random sample consensus: a paradgm for model fttng wth applcatons to mage analyss and automated cartography. Communcatons of the ACM, 24(6): , [6] C. Geyer and K. Danlds. Paracatadoptrc camera calbraton. PAMI, 24(5): , [7] A. Goshtasby. Correcton of mage deformaton from lens dstorton usng bezer patches. Computer Vson, Graphcs, and Image Processng, 47(3): , [8] R. Hartley and S. B. Kang. Parameter-free radal dstorton correcton wth center of dstorton estmaton. PAMI, 29(8): , [9] K. Josephson and M. Byrod. Pose estmaton wth radal dstorton and unknown focal length. In CVPR, [10] Z. Kukelova, M. Bujnak, and T. Pajdla. Real-tme soluton to the absolute pose problem wth unknown radal dstorton and focal length. In ICCV, [11] H. L and R. Hartley. Plane-based calbraton and autocalbraton of a fsh-eye camera. In ACCV [12] T. Sattler, B. Lebe, and L. Kobbelt. Improvng mage-based localzaton by actve correspondence search. In ECCV [13] D. Scaramuzza, A. Martnell, and R. Segwart. A toolbox for easly calbratng omndrectonal cameras. In IROS, [14] J.-P. Tardf, P. Sturm, and S. Roy. Self-calbraton of a general radally symmetrc dstorton model. In ECCV [15] S. Thrthala and M. Pollefeys. Mult-vew geometry of 1d radal cameras and ts applcaton to omndrectonal camera calbraton. In CVPR, [16] S. Thrthala and M. Pollefeys. The radal trfocal tensor: A tool for calbratng the radal dstorton of wde-angle cameras. In CVPR, [17] S. Thrthala and M. Pollefeys. Radal mult-focal tensors. Internatonal Journal of Computer Vson, 96(2): ,

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

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

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

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

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

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

New dynamic zoom calibration technique for a stereo-vision based multi-view 3D modeling system

New dynamic zoom calibration technique for a stereo-vision based multi-view 3D modeling system New dynamc oom calbraton technque for a stereo-vson based mult-vew 3D modelng system Tao Xan, Soon-Yong Park, Mural Subbarao Dept. of Electrcal & Computer Engneerng * State Unv. of New York at Stony Brook,

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

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

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

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

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

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

arxiv: v1 [cs.ro] 8 Jul 2016

arxiv: v1 [cs.ro] 8 Jul 2016 Non-Central Catadoptrc Cameras Pose Estmaton usng 3D Lnes* André Mateus, Pedro Mraldo and Pedro U. Lma arxv:1607.02290v1 [cs.ro] 8 Jul 2016 Abstract In ths artcle we purpose a novel method for planar pose

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

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

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

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

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

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

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

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

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

Image warping and stitching May 5 th, 2015

Image warping and stitching May 5 th, 2015 Image warpng and sttchng Ma 5 th, 2015 Yong Jae Lee UC Davs PS2 due net Frda Announcements 2 Last tme Interactve segmentaton Feature-based algnment 2D transformatons Affne ft RANSAC 3 1 Algnment problem

More information

Geometric Primitive Refinement for Structured Light Cameras

Geometric Primitive Refinement for Structured Light Cameras Self Archve Verson Cte ths artcle as: Fuersattel, P., Placht, S., Maer, A. Ress, C - Geometrc Prmtve Refnement for Structured Lght Cameras. Machne Vson and Applcatons 2018) 29: 313. Geometrc Prmtve Refnement

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

Chapter 6 Programmng the fnte element method Inow turn to the man subject of ths book: The mplementaton of the fnte element algorthm n computer programs. In order to make my dscusson as straghtforward

More information

ROBOT KINEMATICS. ME Robotics ME Robotics

ROBOT KINEMATICS. ME Robotics ME Robotics ROBOT KINEMATICS Purpose: The purpose of ths chapter s to ntroduce you to robot knematcs, and the concepts related to both open and closed knematcs chans. Forward knematcs s dstngushed from nverse knematcs.

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

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

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

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

Helsinki University Of Technology, Systems Analysis Laboratory Mat Independent research projects in applied mathematics (3 cr)

Helsinki University Of Technology, Systems Analysis Laboratory Mat Independent research projects in applied mathematics (3 cr) Helsnk Unversty Of Technology, Systems Analyss Laboratory Mat-2.08 Independent research projects n appled mathematcs (3 cr) "! #$&% Antt Laukkanen 506 R ajlaukka@cc.hut.f 2 Introducton...3 2 Multattrbute

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

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

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

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

AIMS Computer vision. AIMS Computer Vision. Outline. Outline.

AIMS Computer vision. AIMS Computer Vision. Outline. Outline. AIMS Computer Vson 1 Matchng, ndexng, and search 2 Object category detecton 3 Vsual geometry 1/2: Camera models and trangulaton 4 Vsual geometry 2/2: Reconstructon from multple vews AIMS Computer vson

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

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

Lecture 9 Fitting and Matching

Lecture 9 Fitting and Matching In ths lecture, we re gong to talk about a number of problems related to fttng and matchng. We wll formulate these problems formally and our dscusson wll nvolve Least Squares methods, RANSAC and Hough

More information

A Range Image Refinement Technique for Multi-view 3D Model Reconstruction

A Range Image Refinement Technique for Multi-view 3D Model Reconstruction A Range Image Refnement Technque for Mult-vew 3D Model Reconstructon Soon-Yong Park and Mural Subbarao Electrcal and Computer Engneerng State Unversty of New York at Stony Brook, USA E-mal: parksy@ece.sunysb.edu

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

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

Electrical analysis of light-weight, triangular weave reflector antennas

Electrical analysis of light-weight, triangular weave reflector antennas Electrcal analyss of lght-weght, trangular weave reflector antennas Knud Pontoppdan TICRA Laederstraede 34 DK-121 Copenhagen K Denmark Emal: kp@tcra.com INTRODUCTION The new lght-weght reflector antenna

More information

Sequential search. Building Java Programs Chapter 13. Sequential search. Sequential search

Sequential search. Building Java Programs Chapter 13. Sequential search. Sequential search Sequental search Buldng Java Programs Chapter 13 Searchng and Sortng sequental search: Locates a target value n an array/lst by examnng each element from start to fnsh. How many elements wll t need to

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

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

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

A high precision collaborative vision measurement of gear chamfering profile

A high precision collaborative vision measurement of gear chamfering profile Internatonal Conference on Advances n Mechancal Engneerng and Industral Informatcs (AMEII 05) A hgh precson collaboratve vson measurement of gear chamferng profle Conglng Zhou, a, Zengpu Xu, b, Chunmng

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

Wishing you all a Total Quality New Year!

Wishing you all a Total Quality New Year! Total Qualty Management and Sx Sgma Post Graduate Program 214-15 Sesson 4 Vnay Kumar Kalakband Assstant Professor Operatons & Systems Area 1 Wshng you all a Total Qualty New Year! Hope you acheve Sx sgma

More information

Machine Learning: Algorithms and Applications

Machine Learning: Algorithms and Applications 14/05/1 Machne Learnng: Algorthms and Applcatons Florano Zn Free Unversty of Bozen-Bolzano Faculty of Computer Scence Academc Year 011-01 Lecture 10: 14 May 01 Unsupervsed Learnng cont Sldes courtesy of

More information

Finding Intrinsic and Extrinsic Viewing Parameters from a Single Realist Painting

Finding Intrinsic and Extrinsic Viewing Parameters from a Single Realist Painting Fndng Intrnsc and Extrnsc Vewng Parameters from a Sngle Realst Pantng Tadeusz Jordan 1, Davd G. Stork,3, Wa L. Khoo 1, and Zhgang Zhu 1 1 CUNY Cty College, Department of Computer Scence, Convent Avenue

More information

Fitting and Alignment

Fitting and Alignment Fttng and Algnment Computer Vson Ja-Bn Huang, Vrgna Tech Many sldes from S. Lazebnk and D. Hoem Admnstratve Stuffs HW 1 Competton: Edge Detecton Submsson lnk HW 2 wll be posted tonght Due Oct 09 (Mon)

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

For instance, ; the five basic number-sets are increasingly more n A B & B A A = B (1)

For instance, ; the five basic number-sets are increasingly more n A B & B A A = B (1) Secton 1.2 Subsets and the Boolean operatons on sets If every element of the set A s an element of the set B, we say that A s a subset of B, or that A s contaned n B, or that B contans A, and we wrte A

More information

Classifier Selection Based on Data Complexity Measures *

Classifier Selection Based on Data Complexity Measures * Classfer Selecton Based on Data Complexty Measures * Edth Hernández-Reyes, J.A. Carrasco-Ochoa, and J.Fco. Martínez-Trndad Natonal Insttute for Astrophyscs, Optcs and Electroncs, Lus Enrque Erro No.1 Sta.

More information

High-Boost Mesh Filtering for 3-D Shape Enhancement

High-Boost Mesh Filtering for 3-D Shape Enhancement Hgh-Boost Mesh Flterng for 3-D Shape Enhancement Hrokazu Yagou Λ Alexander Belyaev y Damng We z Λ y z ; ; Shape Modelng Laboratory, Unversty of Azu, Azu-Wakamatsu 965-8580 Japan y Computer Graphcs Group,

More information

Array transposition in CUDA shared memory

Array transposition in CUDA shared memory Array transposton n CUDA shared memory Mke Gles February 19, 2014 Abstract Ths short note s nspred by some code wrtten by Jeremy Appleyard for the transposton of data through shared memory. I had some

More information

ESTIMATION OF INTERIOR ORIENTATION AND ECCENTRICITY PARAMETERS OF A HYBRID IMAGING AND LASER SCANNING SENSOR

ESTIMATION OF INTERIOR ORIENTATION AND ECCENTRICITY PARAMETERS OF A HYBRID IMAGING AND LASER SCANNING SENSOR ESTIMATION OF INTERIOR ORIENTATION AND ECCENTRICITY PARAMETERS OF A HYBRID IMAGING AND LASER SCANNING SENSOR A. Wendt a, C. Dold b a Insttute for Appled Photogrammetry and Geonformatcs, Unversty of Appled

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

PROJECTIVE RECONSTRUCTION OF BUILDING SHAPE FROM SILHOUETTE IMAGES ACQUIRED FROM UNCALIBRATED CAMERAS

PROJECTIVE RECONSTRUCTION OF BUILDING SHAPE FROM SILHOUETTE IMAGES ACQUIRED FROM UNCALIBRATED CAMERAS PROJECTIVE RECONSTRUCTION OF BUILDING SHAPE FROM SILHOUETTE IMAGES ACQUIRED FROM UNCALIBRATED CAMERAS Po-Lun La and Alper Ylmaz Photogrammetrc Computer Vson Lab Oho State Unversty, Columbus, Oho, USA -la.138@osu.edu,

More information

Self-Calibration from Image Triplets. 1 Robotics Research Group, Department of Engineering Science, Oxford University, England

Self-Calibration from Image Triplets. 1 Robotics Research Group, Department of Engineering Science, Oxford University, England Self-Calbraton from Image Trplets Martn Armstrong 1, Andrew Zsserman 1 and Rchard Hartley 2 1 Robotcs Research Group, Department of Engneerng Scence, Oxford Unversty, England 2 The General Electrc Corporate

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

Proper Choice of Data Used for the Estimation of Datum Transformation Parameters

Proper Choice of Data Used for the Estimation of Datum Transformation Parameters Proper Choce of Data Used for the Estmaton of Datum Transformaton Parameters Hakan S. KUTOGLU, Turkey Key words: Coordnate systems; transformaton; estmaton, relablty. SUMMARY Advances n technologes and

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

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

Resolving Ambiguity in Depth Extraction for Motion Capture using Genetic Algorithm

Resolving Ambiguity in Depth Extraction for Motion Capture using Genetic Algorithm Resolvng Ambguty n Depth Extracton for Moton Capture usng Genetc Algorthm Yn Yee Wa, Ch Kn Chow, Tong Lee Computer Vson and Image Processng Laboratory Dept. of Electronc Engneerng The Chnese Unversty of

More information

Correspondence-free Synchronization and Reconstruction in a Non-rigid Scene

Correspondence-free Synchronization and Reconstruction in a Non-rigid Scene Correspondence-free Synchronzaton and Reconstructon n a Non-rgd Scene Lor Wolf and Assaf Zomet School of Computer Scence and Engneerng, The Hebrew Unversty, Jerusalem 91904, Israel e-mal: {lwolf,zomet}@cs.huj.ac.l

More information

Model-Based Bundle Adjustment to Face Modeling

Model-Based Bundle Adjustment to Face Modeling Model-Based Bundle Adjustment to Face Modelng Oscar K. Au Ivor W. sang Shrley Y. Wong oscarau@cs.ust.hk vor@cs.ust.hk shrleyw@cs.ust.hk he Hong Kong Unversty of Scence and echnology Realstc facal synthess

More information

Machine Learning 9. week

Machine Learning 9. week Machne Learnng 9. week Mappng Concept Radal Bass Functons (RBF) RBF Networks 1 Mappng It s probably the best scenaro for the classfcaton of two dataset s to separate them lnearly. As you see n the below

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

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

Problem Definitions and Evaluation Criteria for Computational Expensive Optimization

Problem Definitions and Evaluation Criteria for Computational Expensive Optimization Problem efntons and Evaluaton Crtera for Computatonal Expensve Optmzaton B. Lu 1, Q. Chen and Q. Zhang 3, J. J. Lang 4, P. N. Suganthan, B. Y. Qu 6 1 epartment of Computng, Glyndwr Unversty, UK Faclty

More information

Computer Vision. Exercise Session 1. Institute of Visual Computing

Computer Vision. Exercise Session 1. Institute of Visual Computing Computer Vson Exercse Sesson 1 Organzaton Teachng assstant Basten Jacquet CAB G81.2 basten.jacquet@nf.ethz.ch Federco Camposeco CNB D12.2 fede@nf.ethz.ch Lecture webpage http://www.cvg.ethz.ch/teachng/compvs/ndex.php

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

User Authentication Based On Behavioral Mouse Dynamics Biometrics

User Authentication Based On Behavioral Mouse Dynamics Biometrics User Authentcaton Based On Behavoral Mouse Dynamcs Bometrcs Chee-Hyung Yoon Danel Donghyun Km Department of Computer Scence Department of Computer Scence Stanford Unversty Stanford Unversty Stanford, CA

More information

Real-time Joint Tracking of a Hand Manipulating an Object from RGB-D Input

Real-time Joint Tracking of a Hand Manipulating an Object from RGB-D Input Real-tme Jont Tracng of a Hand Manpulatng an Object from RGB-D Input Srnath Srdhar 1 Franzsa Mueller 1 Mchael Zollhöfer 1 Dan Casas 1 Antt Oulasvrta 2 Chrstan Theobalt 1 1 Max Planc Insttute for Informatcs

More information

Outline. Discriminative classifiers for image recognition. Where in the World? A nearest neighbor recognition example 4/14/2011. CS 376 Lecture 22 1

Outline. Discriminative classifiers for image recognition. Where in the World? A nearest neighbor recognition example 4/14/2011. CS 376 Lecture 22 1 4/14/011 Outlne Dscrmnatve classfers for mage recognton Wednesday, Aprl 13 Krsten Grauman UT-Austn Last tme: wndow-based generc obect detecton basc ppelne face detecton wth boostng as case study Today:

More information

TEST-05 TOPIC: OPTICS COMPLETE

TEST-05 TOPIC: OPTICS COMPLETE Q. A boy s walkng under an nclned mrror at a constant velocty V m/s along the x-axs as shown n fgure. If the mrror s nclned at an angle wth the horzontal then what s the velocty of the mage? Y V sn + V

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

Compiler Design. Spring Register Allocation. Sample Exercises and Solutions. Prof. Pedro C. Diniz

Compiler Design. Spring Register Allocation. Sample Exercises and Solutions. Prof. Pedro C. Diniz Compler Desgn Sprng 2014 Regster Allocaton Sample Exercses and Solutons Prof. Pedro C. Dnz USC / Informaton Scences Insttute 4676 Admralty Way, Sute 1001 Marna del Rey, Calforna 90292 pedro@s.edu Regster

More information

NAG Fortran Library Chapter Introduction. G10 Smoothing in Statistics

NAG Fortran Library Chapter Introduction. G10 Smoothing in Statistics Introducton G10 NAG Fortran Lbrary Chapter Introducton G10 Smoothng n Statstcs Contents 1 Scope of the Chapter... 2 2 Background to the Problems... 2 2.1 Smoothng Methods... 2 2.2 Smoothng Splnes and Regresson

More information

The Codesign Challenge

The Codesign Challenge ECE 4530 Codesgn Challenge Fall 2007 Hardware/Software Codesgn The Codesgn Challenge Objectves In the codesgn challenge, your task s to accelerate a gven software reference mplementaton as fast as possble.

More information

The Greedy Method. Outline and Reading. Change Money Problem. Greedy Algorithms. Applications of the Greedy Strategy. The Greedy Method Technique

The Greedy Method. Outline and Reading. Change Money Problem. Greedy Algorithms. Applications of the Greedy Strategy. The Greedy Method Technique //00 :0 AM Outlne and Readng The Greedy Method The Greedy Method Technque (secton.) Fractonal Knapsack Problem (secton..) Task Schedulng (secton..) Mnmum Spannng Trees (secton.) Change Money Problem Greedy

More information

Prof. Feng Liu. Spring /24/2017

Prof. Feng Liu. Spring /24/2017 Prof. Feng Lu Sprng 2017 ttp://www.cs.pd.edu/~flu/courses/cs510/ 05/24/2017 Last me Compostng and Mattng 2 oday Vdeo Stablzaton Vdeo stablzaton ppelne 3 Orson Welles, ouc of Evl, 1958 4 Images courtesy

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

SVO: Semi-Direct Visual Odometry for Monocular and Multi-Camera Systems

SVO: Semi-Direct Visual Odometry for Monocular and Multi-Camera Systems 1 : Sem-Drect Vsual Odometry for Monocular and Mult-Camera Systems Chrstan Forster, Zchao Zhang, Mchael Gassner, Manuel Werlberger, Davde Scaramuzza Abstract Drect methods for Vsual Odometry (VO) have

More information

Calibration of an Articulated Camera System with Scale Factor Estimation

Calibration of an Articulated Camera System with Scale Factor Estimation Calbraton of an Artculated Camera System wth Scale Factor Estmaton CHEN Junzhou, Kn Hong WONG arxv:.47v [cs.cv] 7 Oct Abstract Multple Camera Systems (MCS) have been wdely used n many vson applcatons and

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

R,t. Refractive Structure-from-Motion Through a Flat Refractive Interface

R,t. Refractive Structure-from-Motion Through a Flat Refractive Interface Refractve Structure-from-Moton Through a Flat Refractve Interface Franços Chadebecq,2 Francsco Vasconcelos,2 George Dwyer,2 René Lacher Sébasten Ourseln 2 Tom Vercauteren 2 Danal Stoyanov,2 Surgcal Robot

More information

Semi-Direct Visual Odometry for Monocular, Wide-angle, and Multi-Camera Systems

Semi-Direct Visual Odometry for Monocular, Wide-angle, and Multi-Camera Systems 1 Sem-Drect Vsual Odometry for Monocular, Wde-angle, and Mult-Camera Systems Chrstan Forster, Zchao Zhang, Mchael Gassner, Manuel Werlberger, Davde Scaramuzza Abstract Drect methods for Vsual Odometry

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

ScienceDirect. The Influence of Subpixel Corner Detection to Determine the Camera Displacement

ScienceDirect. The Influence of Subpixel Corner Detection to Determine the Camera Displacement Avalable onlne at www.scencedrect.com ScenceDrect Proceda Engneerng ( ) 8 8 th DAAAM Internatonal Symposum on Intellgent Manufacturng and Automaton, DAAAM The Influence of Subpxel Corner Detecton to Determne

More information

An Approach in Coloring Semi-Regular Tilings on the Hyperbolic Plane

An Approach in Coloring Semi-Regular Tilings on the Hyperbolic Plane An Approach n Colorng Sem-Regular Tlngs on the Hyperbolc Plane Ma Louse Antonette N De Las Peñas, mlp@mathscmathadmueduph Glenn R Lago, glago@yahoocom Math Department, Ateneo de Manla Unversty, Loyola

More information

6.854 Advanced Algorithms Petar Maymounkov Problem Set 11 (November 23, 2005) With: Benjamin Rossman, Oren Weimann, and Pouya Kheradpour

6.854 Advanced Algorithms Petar Maymounkov Problem Set 11 (November 23, 2005) With: Benjamin Rossman, Oren Weimann, and Pouya Kheradpour 6.854 Advanced Algorthms Petar Maymounkov Problem Set 11 (November 23, 2005) Wth: Benjamn Rossman, Oren Wemann, and Pouya Kheradpour Problem 1. We reduce vertex cover to MAX-SAT wth weghts, such that the

More information

Stitching of off-axis sub-aperture null measurements of an aspheric surface

Stitching of off-axis sub-aperture null measurements of an aspheric surface Sttchng of off-axs sub-aperture null measurements of an aspherc surface Chunyu Zhao* and James H. Burge College of optcal Scences The Unversty of Arzona 1630 E. Unversty Blvd. Tucson, AZ 85721 ABSTRACT

More information

Recovering Camera Pose from Omni-directional Images

Recovering Camera Pose from Omni-directional Images Recoveg Camera Pose from Omn-drectonal Images Ada S.K. WAN 1 Angus M.K. SIU 1 Rynson W.H. LAU 1,2 1 Department of Computer Scence, Cty Unversty of Hong Kong, Hong Kong 2 Department of CEIT, Cty Unversty

More information