Reconstruction of Rigid Body Models from Motion Distorted Laser Range Data Using Optical Flow
|
|
- Kelly Oliver
- 5 years ago
- Views:
Transcription
1 Reconstructon of Rgd Body Models from Moton Dstorted Laser Range Data Usng Optcal Flow Eddy Ilg Raner Kümmerle Wolfram Burgard Thomas Brox Abstract The setup of tltng a 2D laser range fnder up and down s a wdespread strategy to acqure 3D pont clouds. Ths setup requres that the scene s statc whle the robot takes a 3D scan. If an object moves through the scene durng the measurement process and one does not take nto account these movements, the resultng model wll get dstorted. Ths paper presents an approach to reconstruct the 3D model of a movng rgd object from the nconsstent set of 2D measurements by the help of a camera. Our approach utlzes optcal flow n the camera mages to estmate the moton n the mage plane and pont-lne constrants to compensate the mssng nformaton about the moton n depth. We combne multple sweeps and/or vews nto to a sngle consstent model usng a pont-to-plane ICP approach and optmze sngle sweeps by smoothng the resultng trajectory. Experments obtaned n real outdoor scenaros wth movng cars demonstrate that our approach yelds accurate models. noddng 2D laser range fnder camera I. INTRODUCTION Compared to stereo cameras, laser range fnders have become popular devces for the acquston of 3D pont clouds n large-scale scenaros because of ther ablty to obtan accurate long-range measurements [1]. A wdely used and cost-effectve setup s to mount a 2D range scanner on a tltng actuator. By noddng the laser scanner ths approach s able to generate accurate models of statc scenes. However, f an object moves durng an entre sweep, whch wth typcal confguratons takes one second or even more, the ndvdual measurements wll not be consstent (see Fgure 1a). In ths paper, we present an approach that utlzes the optcal flow calculated from mages grabbed durng the pont cloud acquston and combnes t wth the laser measurements to estmate the movement of the correspondng object. Accordng to the estmated movement, t then calculates an accurate model of ths object (see Fgures 1b and 1c). The key dea of our approach s to estmate and perform a compensaton for the moton of the movng object. To acheve ths, we employ a vdeo camera that we calbrate to the noddng laser range fnder. As the camera operates at a hgher framerate compared to the 3D pont cloud acquston (1 Hz vs. 1 Hz), our approach estmates the moton of the object n the mage plane from frame to frame usng optcal flow [2] and employs ths nformaton to reproject the ndvdual 2D laser range scanlnes accordng to the movement of the object. Ths, however, requres that we estmate the depth of the ndvdual mage pxels after the moton. To acheve ths, we assume that the perceved object All authors are wth the Unversty of Freburg. Ths work has been partally supported by DFG grant BR 3815/5-1 and the EC under contract numbers ERC LfeNav, FP7-689-ROVINA, and FP EUROPA2. Fg. 1. Top vew of a bulk pont cloud recorded durng 15 sweeps of a noddng 2D laser range fnder of a car movng along the trajectory ndcated by red arrows. The car used as movng object. (c) The colored pont cloud model reconstructed from the dstorted pont cloud above. s rgd. If, at one nstant of tme, we know a partal model of the rgd body, the optcal flow provdes pont-lne constrants for the pose of ths model at another nstant of tme. As a mnmum model we have a non-degenerated sngle scanlne (recorded n one rotaton of the mrror deflectng the laser), whch we assume to be recorded nstantaneously. Trackng the pose of ths model and addng more scanlnes over tme allows for reconstructng the model of the rgd object. Pose trackng wth optcal flow naturally accumulates errors (drft), especally n the vewng drecton of the camera. Therefore, as soon as we record a new sweep of the object, we use the model from ths sweep as a reference and regster the old model to t usng an ICP approach. Ths restrcts drft to the tme wndow between two sweeps. The model recorded durng one sweep s dense n the drecton of the rotaton of the mrror of the laser range fnder, but sparse n the noddng drecton. Due to ths sparseness, the algnment of sweeps s not an easy task. We estmate the underlyng surface of the exstng (possbly denser) model by estmatng ts normals and regster the ponts of the newly recorded sweep by mnmzng ther pont-to-plane dstances to ths surface. To reduce drft, we furthermore constran the moton of the object to the ground plane. Furthermore, we mprove the trackng result by smoothng t wth the algnment determned by ICP. Ths allows us to (c)
2 reduce the nfluence of the drft on the qualty of the model estmated by our approach. Afterwards, we use the smooth trajectory to recalculate the pont cloud of the current sweep before ntegratng t nto the model. II. RELATED WORK The typcal methods to obtan 3D data wth laser range fnders can be roughly dvded nto two categores. The frst operates wth laser range fnders that are rgdly connected to a movng platform, such as a robot [3]. These methods requre a precse estmate of the poston of the robot to obtan an accurate model. The second category actvely actuates the laser range fnder ether by noddng or rotatng t [4], where the laser range fnder tself may have a sngle or multple beams. Note that there are also exceptons. For example, Zbedee by Bosse et al. [5] employs a sprngmounted laser range fnder whch s passvely actuated due to the vbratons nduced by drvng over non-flat surfaces. Obvously, the data of such a 3D scannng devce can be used for varous tasks ncludng object reconstructon or mappng an envronment. In addton to the categores mentoned above, dedcated approaches for model acquston have been proposed. A popular approach to obtan the model s to use a robot for movng the sensor around the object [6], [7]. Krann et al. [8] suggested a method that uses a robotc arm to grasp the object and to move t n front of a depth camera. In contrast to our approach, these methods ether obtan a model of a non-movng object or move the object by themselves, whereas our approach consders the data of a contnuously noddng 2D laser range fnder to obtan a 3D model of a rgd object, whch s movng along an unknown trajectory. Blas et al. [9] present an approach that explctly takes nto account that the scanned object moves. Ther approach teratvely refnes the model but reles on a Lssajous pattern to obtan the range data. Wese et al. [1] propose an approach to correct the dstorton for actve llumnaton stereo n short tme frames. Instead of obtanng a model of an object by scannng a sngle object from multple vews, Ruhnke et al. [11] explot the presence of multple nstances of the objects n the scene to obtan complete 3D models. Ther approach apples spectral clusterng to merge the ndvdual nstances nto one object model. A related problem n computer vson s 3D reconstructon from a movng camera n the presence of ndependently movng objects [12]. Whle a statc scene would come down to a structure-from-moton problem, a movng rgd object requres segmentaton of the object and estmaton of the relatve rgd moton. Our approach combnes the data of a laser range fnder wth vson. Ths combnaton has been appled successfully for other applcatons, such as extendng the range for terran classfcaton [13] or ncreasng the resoluton of the range data [14]. Held et al. [15] apply upsamplng to the sparse range data of a Velodyne scanner. They show that the denser range data leads to better velocty estmates for trackng a movng object. In contrast to our approach, ther method Fg. 2. The calbraton pattern n the camera data and n the remsson data of the laser range fnder. Fg. 3. The deal model of the calbraton pattern. The red dots ndcate the corners of the pattern. The error functon used to detect the rgd n-plane transformaton. Darker areas ndcate lower error. receves a full 3D scan wth 1 Hz, whereas we receve such a scan at only 1 Hz and scanlne-by-scanlne. We explot that the camera operates at a much hgher frame rate than our noddng 2D laser range fnder and track the object n the mage space to account for the dstorton. III. CALIBRATION AND SYNCHRONIZATION A. Calbraton For calbratng the tltng laser range scanner and the camera we employ the calbraton pattern depcted n Fgure 2. To have a dstnct pattern that can be well observed by both the laser range fnder and the camera, the brght areas consst of hghly reflectve materal conductng a scatterng of the laser beam. Ths results n low remsson values (black patches n Fgure 2b) and makes t possble to detect the correspondng areas n the 3D scan. For calbraton, we nod the laser range fnder at a very slow velocty and record multple sweeps to obtan hgh resoluton scans. To establsh a camera projecton matrx P, whch maps the 3D ponts to mage ponts, we are nterested n fndng the set of correspondng patch corners n mage and 3D space. In the mage we compute the ntersectons of the lnes detected from the edges of the patches. In 3D space, we frst ft a plane through the ponts p wth low remsson values (black patches n Fgure 2b) usng a RANSAC approach. Gven ths plane we then nterpolate corners of the patches by fttng an
3 Rotaton Axs r Screw Moton (Twst) p 3 Translaton v p 2 p 1 Fg. 5. Illustraton of a twst ξ. The moton corresponds to a rotaton around the axs ω (red) and a translaton along v (blue) resultng n the screw moton from p 1 to p 3 (green). Rotaton Fg. 4. Calbraton object used to determne the tme shft between the laser range fnder and the camera. Green dots ndcate ponts of the object detected by the laser range fnder due to ther remsson value. The whte arrows ndcate the drecton of movement. Synchronzed laser range fnder and camera data of a movng object. deal model (see Fgure 3a) of the calbraton pattern to the data. To ths end, we defne an error functon E( ) (see Fgure 3b) for the pont locatons. To determne the pose (R α, t) of the pattern on the prevously extracted plane we mnmze α,t N E (R α p + t) (1) =1 usng gradent descent, where R α s a 2D rotaton matrx wth angle α and t s a 2D translaton. Once we know the correspondng locatons of the corners n 3D space and the mage, we establsh a homography [16]. Snce the calbraton object s planar, we take multple rotated vews of the pattern to unquely determne the nternal parameters of the camera and the projecton matrx P. B. Synchronzaton To determne the tme shft between the laser range fnder and the tlt actuator, we explot that objects must have the same coordnates wthn subsequent upward and downward sweeps. Hence, we determne the angle between the ground planes n two subsequent sweeps and teratvely shft the tmestamps of the tlt actuator untl ths angle approaches zero. By movng another calbraton object at hgh veloctes from left to rght, we calbrate the tme shft between the laser and the camera. The surface of ths object (depcted n Fgure 4a) conssts of the same reflectve materal as the calbraton pattern descrbed before. For the synchronzaton, we do not nod the laser range fnder. We detect the pattern n the laser range fnder data by ts low remsson values and project the ponts to mage space (depcted as green dots n Fgure 4). Along the lne (shown n whte) extrapolated from these ponts, we search for nearby pxels wth the color of the calbraton object to determne ts locaton n the mage. For a gven scanlne, we then seek for a camera frame close n tme whch best matches the poston of the object. We store the dfference n tme between the scanlne and the mage and ft a normal dstrbuton to a hstogram over all values. We then use the mean of ths dstrbuton to shft the camera tmestamps to match the laser range fnder tmestamps. Fgure 4b shows a typcal result. IV. MODEL RECONSTRUCTION A. Pont-Lne Correspondences from Optcal Flow To compensate the moton, we assume that a partal model s known at one nstant of tme. In the begnnng, ths s the frst scanlne of the laser range fnder. Let x denote the 3D ponts of ths model. Wth the projecton matrx P from the prevous secton, these ponts can be projected nto the camera mage. We denote the ponts n the mage plane by x. We use large dsplacement optcal flow [2] to compute the dsplacements of x to the next camera frame. The dsplaced mage locaton x = x + restrcts x to the projecton ray of x. 3D lnes can be represented mplctly by so-called Plücker lnes [17]. Let L = (m, n ) be the Plücker lne wth a unt vector n and a moment m that corresponds to the projecton ray x. The Plücker lne representaton then yelds drectly the dstance of an arbtrary 3D pont x to L va [18]: d(l, x) = x n m. (2) Based on these pont-lne constrants, we estmate the sx degrees of freedom of the rgd body moton to move the 3D ponts to ther new poston at the tme when the current camera mage was taken. B. Pose Estmaton For a gven set of non-occluded ponts x we seek for a rgd body transformaton T = (R t) wth a rotaton R and a translaton t that mnmzes d(l, x ). For the purpose of pose estmaton, the twst representaton [19] of T s well suted. As llustrated n Fgure 5, a twst s a screw moton around a rotaton axs ω and a translaton v. Such a twst can be represented as a vector ξ = (ω 1, ω 2, ω 3, v 1, v 2, v 3 ) or n matrx form ˆω = ω 3 ω 2 ( ) ω 3 ω 1 ˆω v, ˆξ = ω 2 ω 1. (3) Multplyng ˆξ by a factor θ allows for an arbtrary scalng of the moton [2]. We can map a scaled twst to a transformaton matrx and back by takng the exponental or logarthm: 1 T = exp(θˆξ), ˆξ = log(t), (4) θ
4 whch can be computed effcently by the Rodrgues formula [19]. Here, we set θ = 1 and seek for a twst that mnmzes the dstance d(l, x ): ξ (exp(ˆξ) π ( x 1 )) 2 n m, (5) where π( ) denotes the projecton from homogeneous to Eucldean coordnates. To reduce drft, we restrct the moton to the ground plane, whch we detect usng RANSAC, by algnng the z-axs of the reference frame to the ground normal and by settng ω 1 =, ω 2 =, and v 3 =. (6) Eq. (5) states a non-lnear least squares problem, whch we solve wth the Gauss-Newton method,.e., we teratvely lnearze exp(ˆξ) I + ˆξ. Hence, Eq. (5) becomes a lnear system n each teraton: ξ ((I + ˆω) x + v) n m 2. (7) We map the twst correspondng to the soluton of ths lnear system to the correspondng transformaton matrx and apply t to the 3D ponts to perform the next teraton. C. Treatment of Outlers Both the estmaton of the optcal flow and the pose of the object are affected by resdual errors. Hence, repeatedly estmatng a new pose gven the prevous result leads to an accumulaton of errors,.e., a drft n the poston and the orentaton of the object. Close to the object boundary, ths can result n model ponts that need to be projected to a background pxel, where the optcal flow does not correspond to the moton of the model. Fgure 6a llustrates such a case. Another cause for such outlers can be local errors of the optcal flow. To be robust to such outlers, we apply robust statstcs and replace the squared error norm n Eq. (5) by a truncated Huber norm [21]. Ths s equvalent to teratvely reweghted least squares. Usng the Gauss-Newton method, we teratvely solve the lnear system of Eq. (7). Let us denote ths system as ξ A ξ b 2. (8) If the correspondence x, x s an outler, the resdual r (ξ) = A ξ b of ths pont s large. To reduce ts nfluence on the soluton ξ, we ntroduce the weght { f r (ξ) > τ w (ξ) =, (9) otherwse 1 r (ξ) +ε whch corresponds to replacng the quadratc norm by the truncated Huber norm wth truncaton at τ. The parameter ɛ lmts the weghts of ponts wth low resdual and, consequently, the condton number of the system matrx. We start wth w = 1 and then solve Eq. (8) by computng ξ k w (ξ k 1 ) A ξ k b 2 (1) n each teraton k. Fgure 6b shows the result of enablng teratvely weghted least squares. Fg. 6. Illustraton of the pose estmaton drft durng the reconstructon of the frst sx sweeps of a car movng to the rght. The area n red shows the optcal flow whle the ponts ndcate the projecton of the model and ts estmated pose. The drft leads to outlers (ndcated by the whte arrows). Applyng Eq. (9) weakens the nfluence of the outlers. Fg. 7. Illustraton of model mergng wth the frst two sweeps of a movng car. Intal msalgnment of the frst two sweeps. The frst sweep s shown n red and the second one n blue. The model after algnng the frst two sweeps wth our approach. One can also see that the sweeps are dense n the drecton of mrror rotaton and sparse n the drecton of noddng. D. Model Mergng Applyng the pose estmaton process descrbed above allows us to track the object and buld ts model by accumulatng ponts. As already mentoned, ths method produces drft. To keep the error of the drft bounded, we buld a new model for each sweep. As the pose of the most recently recorded model s most certan, we algn the formerly tracked model wth the most recent one. Afterwards, we merge the models to an accumulated one, whch s then algned and merged wth the next sweep n the same manner. Let Γ be the accumulated model and let Λ be the model of the most recent sweep. We want to regster both models n a common coordnate frame. To ths end, we consder the trackng result provded by optcal flow as an ntal guess for regsterng both models wth an ICP approach (see Fgure 7a). We, however, have to take nto account that the range data s sparse and that the scanlnes ht the object at dfferent heght levels. Applyng the commonly used pontto-pont metrc would algn the dense scanlnes and yeld sub-optmal results. Thus, we nstead consder a pont-toplane metrc. Consequently, we have to estmate the normal vectors. Snce the qualty of the estmate of a normal vector depends on the densty of the underlyng pont cloud, we compute the normal vectors on the accumulated model Γ, whch s n general denser than the newly recorded model Λ. To estmate the normal vector n of a pont p Γ, we
5 Tn 1 yn TA y1 y1 Fg. 8. The trajectory optmzaton. Image shows the pose estmates between two specfc poses y1 and yn. Blue crcles ndcate the estmated poses drectly after a laser sweep was completed. Red lnes ndcate pose updates due to optcal flow pose estmaton whle blue lnes ndcate pose updates due to an ICP algnment. Correspondng optmzed trajectory. frst fnd the neghbors wthn a certan radus. We compute the normal vector n as the Egenvector correspondng to the smallest Egenvalue of the covarance matrx of these neghborng ponts. Addtonally, we determne the sgn of the surface normal such that t ponts towards the camera. We furthermore assume that the surfaces of the captured objects are convex. Ths allows us to detect occlusons when the object s movng by dentfyng ponts whose normals do no longer pont n the drecton of the camera. Specfcally, we mark ponts as occluded f the angle between the vector from the pont to the camera and the normal exceeds 6. We do not consder occluded ponts n the pose estmaton step. A representaton of the tangental plane to a pont p wth normal n s n x d =, (11) where d = n p. Furthermore, n x d represents the dstance of a pont x to ths plane. We teratvely seek for the nearest neghbors q Λ to p bounded by a maxmum dstance dmax and algn Λ to Γ by mnmzng X 2 n (TA q ) d. (12) TA Representng TA as a twst (see Eq. (4)) allows us to apply the constrant of Eq. (6), namely that the object moves on the ground plane. Agan, we fnd the soluton of Eq. (12) usng the Gauss-Newton method analogously to the pose estmaton problem descrbed before. Fgure 7b depcts the outcome of our regstraton approach. We found that the robustness of the algnment of the sparse clouds heavly depends on the parameter dmax. In contrast to other ICP mplementatons, our approach starts wth dmax = 1 cm n each teraton and only ncreases dmax f the number of neghbors found s nsuffcent (less than one-thrd of the sze of Λ). E. Trajectory Optmzaton As mentoned above, the pose of the reconstructed object s more accurate after we regstered our exstng model Γ and the pont cloud Λ of the current sweep. The algnment step corresponds to a correcton of the error accumulated by channg up the pose estmates (T1,..., Tn 1 ) from optcal flow. Fgure 8a llustrates such a chan along wth the ICP algnment that yelds the transformaton matrx TA. To account for the pose estmaton errors made durng the constructon of Λ, we perform a smoothng of the trajectory to obtan the maxmum lkelhood estmate for the trajectory (c) Fg. 9. Pctures and pont clouds of the dfferent objects: Aud TT, Van, (c) Polo. RMS [cm] T1 yn Aud TT Van Polo A1 A3 A2 Aud TT V1 V2 Van V3 P1 P2 Polo P3 Fg. 1. RMS errors of the dfferent datasets from Fgure 11 compared to the dfferent ground truths from Fgure 9. As vsble, the RMS to the correct ground truth s always the smallest and therefore allows for a classfcaton. of the object. Let (y1,..., yn ) be the poses of the movng object whle Λ s acqured. Addtonally, let e(y, yj, T) be an error functon whch computes the dfference between a gven estmated transformaton T and the expected value gven the current state of the poses y and yj. We determne y1,...,n by solvng ke(y1, yn, TA )k2ψ + y1,...,n n 1 X ke(y, y+1, T )k2σ, (13) =1 where k k2σ s the squared Mahalanobs dstance weghted by the covarance matrx Σ. Here, Ψ and Σ allow us to balance the nfluence of the measurements. Agan, we solve Eq. (13) wth Gauss-Newton as mplemented n g2 o [22]. Ths leads to a smoothed trajectory y1,...,n (see Fgure 8b) that we use to rebuld the model acqured durng the most recent sweep. V. R ESULTS We evaluate our approach on several real-world data sets n whch three dfferent cars are movng through the scene. A Hokuyo UTM-3LX laser rotatng wth 4 Hz mounted on a servo that s tltng from -5 to +3 provded the 3D range measurements. Addtonally, we recorded the mage data of one camera of a Pont Grey Bumblebee2 stereo camera runnng at 1 Hz and a resoluton of pxels. We use the mplementaton from Brox and Malk [2] for computng the optcal flow. For a quanttatve evaluaton, we acqured dense pont clouds of the stll standng vehcles depcted n Fgure 9 by manually steerng a robot around the vehcle. We appled sparse surface adjustment [23] to jon the vews and to obtan ground truth models. To assess the qualty of our model reconstructon from a movng vehcle, we constran the models to the ground plane and algn the reconstructon to the ground truth wth the prevously ntroduced pont-toplane ICP approach. As an ntalzaton we algn the four contact ponts of the wheels. For each model pont we then
6 Aud Aud TT TT - A1 - A Sweeps Aud Aud TT TT - A1 - A Sweeps Aud Aud TT TT- A1 - A Sweeps Aud Van TT - V1 A Sweeps Aud Van TT - V2 A Sweeps Aud Van TT - V3 A Sweeps Aud Polo TT - A1 P1 141 Sweeps Aud Polo TT - P2 A Sweeps Aud Polo TT - A1 P Sweeps Fg. 11. Illustraton of the dfferent datasets. The top row shows the trajectory of the movng object estmated by our approach along wth a top vew of the bulk pont cloud. The bottom row llustrates the models reconstructed by our approach. fnd the nearest neghbor n the ground truth and compute the root mean squared error (RMS). We not only compute the RMS of a vehcle compared to tself but also to the other models. The results are shown n Fgure 1. For the evaluaton we recorded three dfferent moton scenes of dfferent lengths (from 3 to 3 seconds) for each vehcle, as llustrated n Fgure 11. The trajectores nclude lnear motons towards the camera, from left-to-rght and rght-to-left, stops, curves and crcles. For each vehcle, n case of a crcle, we are able to reconstruct a 36 model. The RMS wth respect to the correspondng ground truth model typcally les between 3 cm and 9 cm. As vsble from Fgure 1, n all cases (even wth only 3 sweeps), the RMS of the reconstructed models suffces to dstngush between dfferent vehcles wthout the use of further features. Ths ndcates that we can use our approach to classfy movng objects based on ther 3D shape nstead of relyng on the vsual appearance alone. VI. CONCLUSION In ths paper, we propose a method to reconstruct models of movng rgd objects captured by a noddng laser range fnder and a vdeo camera. To ths end, our approach consders optcal flow to track the object n the camera frame and to correct the dstorton n the range data nduced by the moton of the object. The results from dfferent moton scenes of vehcles ndcate that our approach allows us to obtan accurate reconstructons. Our evaluaton shows that reconstructon errors le n the range of 3 cm to 9 cm. Furthermore, one could use our approach to classfy the movng object based on correspondng, prevously recorded 3D models. There are several optons for future work. Potental extensons regard a real-tme mplementaton or the relaxaton of the ground plane or rgdty assumpton. REFERENCES [1] S. Kumar, D. Gupta, and S. Yadav, Sensor fuson of laser and stereo vson camera for depth estmaton and obstacle avodance, Internatonal Journal of Computer Applcatons, vol. 1, no. 26, 21. [2] T. Brox and J. Malk, Large dsplacement optcal flow: descrptor matchng n varatonal moton estmaton, IEEE Transactons on Pattern Analyss and Machne Intellgence, vol. 33, pp , 211. [3] G. Sbley, C. Me, I. Red, and P. Newman, Vast scale outdoor navgaton usng adaptve relatve bundle adjustment, Int. Journal of Robotcs Research, vol. 29, no. 8, pp , July 21. [4] O. Wulf and B. Wagner, Fast 3D-scannng methods for laser measurement systems, n Int. Conf. on Control Systems and Computer Scence (CSCS), 23. [5] M. Bosse, R. Zlot, and P. Flck, Zebedee: Desgn of a sprng-mounted 3-D range sensor wth applcaton to moble mappng, IEEE Trans. on Robotcs, vol. 28, no. 5, pp , 212. [6] R. Trebel, B. Frank, J. Meyer, and W. Burgard, Frst steps towards a robotc system for flexble volumetrc mappng of ndoor envronments, Proc. of the IFAC/EURON Symposum on Intellgent Autonomous Vehcles (IAV), 24. [7] T. Fossotte, O. Stasse, A. Escande, P.-B. Weber, and A. Kheddar, A two-steps next-best-vew algorthm for autonomous 3D object modelng by a humanod robot, n Proc. of the IEEE Int. Conf. on Robotcs & Automaton (ICRA), 29. [8] M. Krann, B. Curless, and D. Fox, Autonomous generaton of complete 3D object models usng next best vew manpulaton plannng, n Proc. of the IEEE Int. Conf. on Robotcs & Automaton (ICRA), 211. [9] F. Blas, M. Pcard, and G. Godn, Accurate 3D acquston of freely movng objects, n Proc. of the 2nd Int. Symposum on 3D Data Processng, Vsualzaton and Transmsson, 24. [1] T. Wese, B. Lebe, and L. Van Gool, Fast 3D scannng wth automatc moton compensaton, n Proc. of the IEEE Conf. on Comp. Vson and Pattern Recognton (CVPR), 27. [11] M. Ruhnke, B. Steder, G. Grsett, and W. Burgard, Unsupervsed learnng of 3D object models from partal vews, n Proc. of the IEEE Int. Conf. on Robotcs & Automaton (ICRA), 29. [12] K. E. Ozden, K. Schndler, and L. Van Gool, Multbody structurefrom-moton n practce, IEEE Transactons on Pattern Analyss and Machne Intellgence, vol. 32, no. 6, pp , 21. [13] H. Dahlkamp, A. Kaehler, D. Stavens, S. Thrun, and G. Bradsk, Selfsupervsed monocular road detecton n desert terran, n Proc. of Robotcs: Scence and Systems (RSS), 26. [14] J. Dolson, J. Baek, C. Plagemann, and S. Thrun, Upsamplng range data n dynamc envronments, n Proc. of the IEEE Conf. on Comp. Vson and Pattern Recognton (CVPR), 21. [15] D. Held, J. Levnson, and S. Thrun, Precson trackng wth sparse 3D and dense color 2D data, n Proc. of the IEEE Int. Conf. on Robotcs & Automaton (ICRA), 213. [16] Z. Zhang, A flexble new technque for camera calbraton, IEEE Trans. on Pattern Analyss and Machne Intellgence (PAMI), vol. 22, no. 11, pp , 2. [17] F. Shevln, Analyss of orentaton problems usng Plücker lnes. n Internatonal Conference on Pattern Recognton (ICPR), vol. 1, Brsbane, 1998, pp [18] T. Brox, B. Rosenhahn, J. Gall, and D. Cremers, Combned regonand moton-based 3D trackng of rgd and artculated objects, IEEE Transactons on Pattern Analyss and Machne Intellgence, vol. 32, no. 3, pp , 21. [19] R. M. Murray, Z. L, and S. S. Sastry, Mathematcal Introducton to Robotc Manpulaton. Baton Rouge: CRC Press, [2] B. Rosenhahn, T. Brox, and H.-P. Sedel, Scaled moton dynamcs for markerless moton capture, n Internatonal Conference on Computer Vson and Pattern Recognton (CVPR), 27. [21] P. J. Huber, Robust Statstcs. New York: Wley, [22] R. Kümmerle, G. Grsett, H. Strasdat, K. Konolge, and W. Burgard, g2o: A general framework for graph optmzaton, n Proc. of the IEEE Int. Conf. on Robotcs & Automaton (ICRA), 211. [23] M. Ruhnke, R. Kümmerle, G. Grsett, and W. Burgard, Hghly accurate 3D surface models by sparse surface adjustment, n Proc. of the IEEE Int. Conf. on Robotcs & Automaton (ICRA), 212.
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 informationStructure 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 informationImprovement 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 informationCS 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 informationSLAM 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 informationImage 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 informationFitting & 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 informationAn 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 informationFeature 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 informationFace 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 informationOutline. 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 informationMOTION 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 informationActive 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 informationMulti-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 informationMULTISPECTRAL 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 informationOutline. 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 informationHermite 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 informationMulti-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 informationNew 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 informationy 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 informationLearning 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 informationReal-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 informationFitting: 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 informationFitting 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 informationSimultaneous 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 informationSupport 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 informationSkew 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 informationA 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 informationEcient 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 informationMETRIC ALIGNMENT OF LASER RANGE SCANS AND CALIBRATED IMAGES USING LINEAR STRUCTURES
METRIC ALIGNMENT OF LASER RANGE SCANS AND CALIBRATED IMAGES USING LINEAR STRUCTURES Lorenzo Sorg CIRA the Italan Aerospace Research Centre Computer Vson and Vrtual Realty Lab. Outlne Work goal Work motvaton
More informationLECTURE : 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 informationProf. 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 informationCorner-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 informationLine-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 informationProper 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 informationReal-time Motion Capture System Using One Video Camera Based on Color and Edge Distribution
Real-tme Moton Capture System Usng One Vdeo Camera Based on Color and Edge Dstrbuton YOSHIAKI AKAZAWA, YOSHIHIRO OKADA, AND KOICHI NIIJIMA Graduate School of Informaton Scence and Electrcal Engneerng,
More informationUnsupervised 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 informationA 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 informationVideo Object Tracking Based On Extended Active Shape Models With Color Information
CGIV'2002: he Frst Frst European Conference Colour on Colour n Graphcs, Imagng, and Vson Vdeo Object rackng Based On Extended Actve Shape Models Wth Color Informaton A. Koschan, S.K. Kang, J.K. Pak, B.
More informationA 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 informationMulti-View Face Alignment Using 3D Shape Model for View Estimation
Mult-Vew Face Algnment Usng 3D Shape Model for Vew Estmaton Yanchao Su 1, Hazhou A 1, Shhong Lao 1 Computer Scence and Technology Department, Tsnghua Unversty Core Technology Center, Omron Corporaton ahz@mal.tsnghua.edu.cn
More informationThe 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 informationReducing 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 informationS1 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 informationExploiting Building Information from Publicly Available Maps in Graph-Based SLAM
Explotng Buldng Informaton from Publcly Avalable Maps n Graph-Based SLAM Olga Vysotska Cyrll Stachnss Abstract Maps are an mportant component of most robotc navgaton systems and buldng maps under uncertanty
More informationAccounting 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 informationUser 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 informationImage Fusion With a Dental Panoramic X-ray Image and Face Image Acquired With a KINECT
Image Fuson Wth a Dental Panoramc X-ray Image and Face Image Acqured Wth a KINECT Kohe Kawa* 1, Koch Ogawa* 1, Aktosh Katumata* 2 * 1 Graduate School of Engneerng, Hose Unversty * 2 School of Dentstry,
More informationHierarchical Optimization on Manifolds for Online 2D and 3D Mapping
Herarchcal Optmzaton on Manfolds for Onlne 2D and 3D Mappng Gorgo Grsett Raner Kümmerle Cyrll Stachnss Udo Frese Chrstoph Hertzberg Abstract In ths paper, we present a new herarchcal optmzaton soluton
More informationParallelism 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 informationDynamic wetting property investigation of AFM tips in micro/nanoscale
Dynamc wettng property nvestgaton of AFM tps n mcro/nanoscale The wettng propertes of AFM probe tps are of concern n AFM tp related force measurement, fabrcaton, and manpulaton technques, such as dp-pen
More informationArticulated Motion Capture from Visual Hulls in High Dimensional Configuration Spaces
Semnar Presentaton June 18th, 010 Artculated Moton Capture from Vsual Hulls n Hgh Dmensonal Confguraton Spaces Azawa Yamasak Lab D, 48-09741, 羅衛蘭 ABSTRACT In ths paper, we propose a novel approach for
More informationComputer 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 information3D 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 informationSIMULTANEOUS 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 informationComputer Vision I. Xbox Kinnect: Rectification. The Fundamental matrix. Stereo III. CSE252A Lecture 16. Example: forward motion
Xbox Knnect: Stereo III Depth map http://www.youtube.com/watch?v=7qrnwoo-8a CSE5A Lecture 6 Projected pattern http://www.youtube.com/watch?v=ceep7x-z4wy The Fundamental matrx Rectfcaton The eppolar constrant
More informationDetection 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 informationA 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 informationA 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 informationGeometric 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 informationOptimizing Document Scoring for Query Retrieval
Optmzng Document Scorng for Query Retreval Brent Ellwen baellwe@cs.stanford.edu Abstract The goal of ths project was to automate the process of tunng a document query engne. Specfcally, I used machne learnng
More information12/2/2009. 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 informationResolving 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 informationNUMERICAL 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 informationCompiler 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 informationAn 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 informationAngle-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 informationPROJECTIVE 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 informationA 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 informationMobile Robot Localization and Mapping by Scan Matching using Laser Reflection Intensity of the SOKUIKI Sensor
Moble Robot Localzaton and Mappng by Scan Matchng usng Reflecton Intensty of the SOKUIKI Sensor *HARA Yoshtaka, KAWATA Hrohko, OHYA Akhsa, YUTA Shn ch Intellgent Robot Laboratory Unversty of Tsukuba 1-1-1,
More informationRange Data Registration Using Photometric Features
Range Data Regstraton Usng Photometrc Features Joon Kyu Seo, Gregory C. Sharp, and Sang Wook Lee Dept. of Meda Technology, Sogang Unversty, Seoul, Korea Dept. of Radaton Oncology, Massachusetts General
More informationUnsupervised Learning and Clustering
Unsupervsed Learnng and Clusterng Supervsed vs. Unsupervsed Learnng Up to now we consdered supervsed learnng scenaro, where we are gven 1. samples 1,, n 2. class labels for all samples 1,, n Ths s also
More informationA 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 informationGeneralized-ICP. Aleksandr V. Segal Stanford University Dirk Haehnel Stanford University
Generalzed-ICP Aleksandr V. Segal Stanford Unversty Emal: avsegal@cs.stanford.edu Drk Haehnel Stanford Unversty Emal: haehnel@stanford.edu Sebastan hrun Stanford Unversty Emal: thrun@stanford.edu Abstract
More informationEfficient Multi-View Object Recognition and Full Pose Estimation
Effcent Mult-Vew Object Recognton and Full Pose Estmaton Alvaro Collet Sddhartha S. Srnvasa Abstract We present an approach for effcently recognzng all objects n a scene and estmatng ther full pose from
More informationHigh resolution 3D Tau-p transform by matching pursuit Weiping Cao* and Warren S. Ross, Shearwater GeoServices
Hgh resoluton 3D Tau-p transform by matchng pursut Wepng Cao* and Warren S. Ross, Shearwater GeoServces Summary The 3D Tau-p transform s of vtal sgnfcance for processng sesmc data acqured wth modern wde
More informationA Volumetric Approach for Interactive 3D Modeling
A Volumetrc Approach for Interactve 3D Modelng Dragan Tubć Patrck Hébert Computer Vson and Systems Laboratory Laval Unversty, Ste-Foy, Québec, Canada, G1K 7P4 Dens Laurendeau E-mal: (tdragan, hebert, laurendeau)@gel.ulaval.ca
More informationLobachevsky 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 informationCORRELATION 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 informationHuman Skeleton Reconstruction for Optical Motion Capture
Journal of Computatonal Informaton Systems 9: 0 (013) 8073 8080 Avalable at http://www.jofcs.com Human Skeleton Reconstructon for Optcal Moton Capture Guanghua TAN, Melan ZHOU, Chunmng GAO College of Informaton
More informationMultiple Frame Motion Inference Using Belief Propagation
Multple Frame Moton Inference Usng Belef Propagaton Jang Gao Janbo Sh The Robotcs Insttute Department of Computer and Informaton Scence Carnege Mellon Unversty Unversty of Pennsylvana Pttsburgh, PA 53
More informationClassifier Swarms for Human Detection in Infrared Imagery
Classfer Swarms for Human Detecton n Infrared Imagery Yur Owechko, Swarup Medasan, and Narayan Srnvasa HRL Laboratores, LLC 3011 Malbu Canyon Road, Malbu, CA 90265 {owechko, smedasan, nsrnvasa}@hrl.com
More informationWishing 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 informationCS434a/541a: Pattern Recognition Prof. Olga Veksler. Lecture 15
CS434a/541a: Pattern Recognton Prof. Olga Veksler Lecture 15 Today New Topc: Unsupervsed Learnng Supervsed vs. unsupervsed learnng Unsupervsed learnng Net Tme: parametrc unsupervsed learnng Today: nonparametrc
More informationFEATURE 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 informationPHOTOGRAMMETRIC ANALYSIS OF ASYNCHRONOUSLY ACQUIRED IMAGE SEQUENCES
PHOTOGRAMMETRIC ANALYSIS OF ASYNCHRONOUSLY ACQUIRED IMAGE SEQUENCES Karsten Raguse 1, Chrstan Hepke 2 1 Volkswagen AG, Research & Development, Dept. EZTV, Letter Box 1788, 38436 Wolfsburg, Germany Emal:
More informationLarge 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 informationCollaboratively Regularized Nearest Points for Set Based Recognition
Academc Center for Computng and Meda Studes, Kyoto Unversty Collaboratvely Regularzed Nearest Ponts for Set Based Recognton Yang Wu, Mchhko Mnoh, Masayuk Mukunok Kyoto Unversty 9/1/013 BMVC 013 @ Brstol,
More informationSubspace 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 information3D Rigid Facial Motion Estimation from Disparity Maps
3D Rgd Facal Moton Estmaton from Dsparty Maps N. Pérez de la Blanca 1, J.M. Fuertes 2, and M. Lucena 2 1 Department of Computer Scence and Artfcal Intellgence ETSII. Unversty of Granada, 1871 Granada,
More informationA Background Subtraction for a Vision-based User Interface *
A Background Subtracton for a Vson-based User Interface * Dongpyo Hong and Woontack Woo KJIST U-VR Lab. {dhon wwoo}@kjst.ac.kr Abstract In ths paper, we propose a robust and effcent background subtracton
More informationarxiv: v1 [cs.ro] 21 Jul 2018
Mult-sesson Map Constructon n Outdoor Dynamc Envronment* Xaqng Dng 1, Yue Wang 1, Huan Yn 1, L Tang 1, Rong Xong 1 arxv:1807.08098v1 [cs.ro] 21 Jul 2018 Abstract Map constructon n large scale outdoor envronment
More informationPose Estimation in Heavy Clutter using a Multi-Flash Camera
2010 IEEE Internatonal Conference on Robotcs and Automaton Anchorage Conventon Dstrct May 3-8, 2010, Anchorage, Alaska, USA Pose Estmaton n Heavy Clutter usng a Mult-Flash Camera Mng-Yu Lu, Oncel Tuzel,
More informationLocal Quaternary Patterns and Feature Local Quaternary Patterns
Local Quaternary Patterns and Feature Local Quaternary Patterns Jayu Gu and Chengjun Lu The Department of Computer Scence, New Jersey Insttute of Technology, Newark, NJ 0102, USA Abstract - Ths paper presents
More informationTN348: 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 informationImproved SIFT-Features Matching for Object Recognition
Improved SIFT-Features Matchng for Obect Recognton Fara Alhwarn, Chao Wang, Danela Rstć-Durrant, Axel Gräser Insttute of Automaton, Unversty of Bremen, FB / NW Otto-Hahn-Allee D-8359 Bremen Emals: {alhwarn,wang,rstc,ag}@at.un-bremen.de
More informationLoad Balancing for Hex-Cell Interconnection Network
Int. J. Communcatons, Network and System Scences,,, - Publshed Onlne Aprl n ScRes. http://www.scrp.org/journal/jcns http://dx.do.org/./jcns.. Load Balancng for Hex-Cell Interconnecton Network Saher Manaseer,
More informationObject Recognition Based on Photometric Alignment Using Random Sample Consensus
Vol. 44 No. SIG 9(CVIM 7) July 2003 3 attached shadow photometrc algnment RANSAC RANdom SAmple Consensus Yale Face Database B RANSAC Object Recognton Based on Photometrc Algnment Usng Random Sample Consensus
More informationLecture 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 informationClassifying Acoustic Transient Signals Using Artificial Intelligence
Classfyng Acoustc Transent Sgnals Usng Artfcal Intellgence Steve Sutton, Unversty of North Carolna At Wlmngton (suttons@charter.net) Greg Huff, Unversty of North Carolna At Wlmngton (jgh7476@uncwl.edu)
More information