Accurate Object Localization in 3D Laser Range Scans
|
|
- Daniela Green
- 5 years ago
- Views:
Transcription
1 Point Distibution numbe of points distances [cm] 7 Accuate Object Localization in 3D Lase Range Scans Andeas Nüchte, Kai Lingemann, Joachim Hetzbeg Univesity of Osnabück, Institute fo Compute Science Knowledge-Based Systems Reseach Goup Albechtstaße 28 D Osnabück, Gemany {nuechte lingemann hetzbeg}@infomatik.uni-osnabueck.de Hatmut Sumann Faunhofe Institute fo Autonomous Intelligent Systems (AIS) Schloss Bilinghoven D Sankt Augustin, Gemany hatmut.sumann@ais.faunhofe.de Abstact This pape pesents a novel method fo object detection and classification in 3D lase ange data that is acquied by an autonomous mobile obot. Unesticted objects ae leaned using classification and egession tees (CARTs) and using an Ada Boost leaning pocedue. Off-sceen endeed depth and eflectance images seve as an input fo leaning. The pefomance of the classification is impoved by combining both senso modalities, which ae independent fom extenal light. This enables highly accuate, fast and eliable 3D object localization with point matching. Competitive leaning is used fo evaluating the accuacy of the object localization. I. INTRODUCTION Envionment peception is a basic poblem in the design of autonomous mobile cognitive systems, i.e., of a mobile obot. A cucial pat of the peception is to lean, detect, localize and ecognize objects, which has to be done with limited esouces. The pefomance of such a obot highly depends on the accuacy and eliability of its pecepts and on the computational effot of the involved intepetation pocess. Pecise localization of objects is the all-dominant step in any navigation o manipulation task. This pape poposes a new method fo the leaning, fast detection and localization of instances of 3D object classes. The appoach uses 3D lase ange and eflectance data acquied by an autonomous mobile obot to peceive the 3D objects. The 3D ange and eflectance data ae tansfomed into images by off-sceen endeing. Based on the ideas of Viola and Jones [25], we built a cascade of classifies, i.e., a linea decision tee. The classifies ae composed of classification and egession tees (CARTs) and model the objects with thei view dependencies. Each CART makes its decisions based on featue classifies and leaned etun values. The featues ae edge, line, cente suound, o otated featues. Lienhat et. al and Viola and Jones have implemented a method fo computing effectivly these featues using an intemediate epesentation, namely, integal image [12], [25]. Fo leaning object classes, a boosting technique, paticulaly, Ada Boost, is used [6]. Afte detection, the object is localized using a matching technique. Heeby the pose is detemined with six degees of feedom, i.e., with espect to the x, y, and z positions and the oll, yaw and pitch angles. Finally the quality of the object localization is evaluated by fast subsampling of the scanned 3D data. The esulting appoach fo object detection is eliable and eal-time capable and combines ecent esults in compute vision with the emeging technology of 3D lase scannes. Fig. 1 gives an oveview of the implemented system. II. STATE OF THE ART Common appoaches of object detection use infomation of CCD-cameas that povide a view of the obot s envionment. Nevetheless, cameas ae difficult to use in natual envionments with changing light conditions. Robot contol achitectues that include obot vision mainly ely on tacking, e.g., distinctive, local, scale invaiant featues [18], light souces [11] o the ceilings [5]. Othe camea-based appoaches to obot vision, e.g., steeo cameas and stuctue fom motion, have difficulties poviding navigation infomation fo a mobile obot in eal-time. Camea-based systems have poblems localizing objects pecisely, i.e., single cameas estimate the object distance only oughly using the known object size due to the esolution. Estimating depth with steeo is impecise eithe: Fo obots, the width of the steeo base line is limited 3D Scan Off Sceen Classification Ray Tacing Model Matching Evaluation Result endeing Visualization depth efl. depth efl. compaing subsampled 3D models "... ///// database Fig. 1. System Oveview. Afte acquiing 3D scans, depth and eflection images ae geneated. In these images, objects ae detected using a leaned epesentation fom a database. Ray tacing selects the points coesponding to the 2D pojection of the object. A 3D model is matched into these points, followed by an evaluation step.
2 to small values (e.g., < 20 cm), esulting in a typical z-axis eo of about 78 cm fo objects at the scanne s maximum anging distance of about 8 m. Many cuent successful obots ae equipped with distance sensos, mainly 2D lase ange findes [24]. 2D scannes cannot detect 3D obstacles outside thei scan plane. Cuently a geneal tend exists to use 3D lase ange findes and build 3D maps [2], [19], [22], [23]. Nevetheless, only little wok has been done in intepeting the obtained 3D models. In [14] we show how complete scenes, made of seveal automatically egisteed 3D scans, ae labeled using elations given in a semantic net. Object detection in 3D lase scans fom mobile obots was pesented in [13]. This appoach is extended hee: Fist, CARTs ae used fo a moe sophisticated object detection, second, objects ae localized in 3D space using point based matching, and thid, the accuacy of the matching is evaluated. In the aea of object ecognition and classification in 3D ange data, Johnson and Hebet use the well-known ICP algoithm [4] fo egisteing 3D shapes into a common coodinate system [10]. The necessay initial guess of the ICP algoithm is done by detecting the object with spin images [10]. This appoach was extended by Shapio et al. [16]. In contast to ou poposed method, both appoaches use local, memoy consuming suface signatues based on pio ceated mesh epesentations of the objects. Futhemoe, spin images ae not able to model complicated objects, i.e., objects with nonsmooth, o non-poducible mesh epesentation. One of the objects used in this pape, the volksbot [1], is of such a stuctue (Fig. 9). Besides spin images, seveal suface epesentation schemes ae in use fo computing an initial alignment. Stein and Medioni pesented the notion of splash to epesent the nomals along a geodesic cicle of a cente point, which is the local Gauss map fo 3D object ecognition with a database [20]. Ashock et al. poposed a paiwise geometic histogam to find coesponding facets between two sufaces that ae epesented by tiangle meshes [3]. Hamonic maps and thei use in suface matching have been used by Zhang and Hebet [26]. Recently, Sun and colleagues have suggested so-called point fingepints : They compute a set of 2D contous that ae pojections of geodesic cicles onto the tangent plane and compute similaities between them [21]. All these appoaches take the local geomety of the sufaces into account, i.e., meshes. They have poblems coping with unstuctued point clouds. III. THE AUTONOMOUS MOBILE ROBOT KURT3D A. The Kut Robot Platfom Kut3D (Fig. 2) is a mobile obot platfom with a size of 45 cm (length) 33 cm (width) 26 cm (height) and a weight of 15.6 kg, both indoo as well as outdoo models exist. Equipped with the 3D lase ange finde, the height inceases to 47 cm and the weight inceases to 22.6 kg. 1 1 Videos of the exploation with the autonomous mobile obot can be found at: Fig. 2. The autonomous mobile obot Kut3D equipped with the 3D lase ange finde as pesented at RoboCup The scannes technical basis is a SICK 2D lase ange finde (LMS-200). Kut3D s maximum velocity is 5.2 m/s (autonomously contolled: 4.0 m/s). Two 90W motos ae used to powe the 6 wheels. Compaed to the oiginal Kut3D obot platfom, this cuent vesion has lage wheels, whee the middle wheels ae shifted outwads. Font and eal wheels have no tead patten to enhance otating. Kut3D opeates fo about 4 hous with one battey (28 NiMH cells, capacity: 4500 mah) chage. The coe of the obot is an Intel-Centino-1400 MHz with 768 MB RAM and a Linux opeating system. An embedded 16-Bit CMOS micocontolle is used to contol the moto. B. The 3D Lase Scanne The 3D lase ange finde (Fig. 2) is built on the basis of a 2D ange finde by extension with a mount and a standad sevo moto [23]. The 2D lase ange finde is attached to the mount in the cente of otation fo achieving a contolled pitch motion. The sevo is connected on the left side (Fig. 2). The 3D lase scanne opeates up to 5h (Scanne: 17 W, 20 NiMH cells with a capacity of 4500 mah, Sevo: 0.85 W, 4.5 V with batteies of 4500 mah) on one battey pack. IV. DETECTING OBJECTS IN 3D LASER DATA A. Rendeing Images fom Scan Data Afte scanning, the 3D data points ae pojected by an offsceen OpenGL-based endeing module onto an image plane to ceate 2D images. The camea fo this pojection is located in the lase souce, thus all points ae unifomly distibuted and enlaged to emove gaps between them on the image plane. Fig. 9 shows eflectance images and endeed depth images (distances encoded by gey-values) as well as point clouds. B. Featue Detection using Integal Images Thee ae many motivations fo using featues athe than pixels diectly. Fo mobile obots, a citical motivation is that featue-based systems opeate much faste than pixel-based systems [25]. The featues used hee have the same stuctue as the Haa basis functions, i.e., step functions intoduced by Alfed Haa to define wavelets [8]. They ae also used in [12], [15], [25]. Fig. 3 (left) shows the eleven basis featues, i.e.,
3 ============<<<<<<<<<<< edge, line, diagonal and cente suound featues. The base esolution of the object detecto is fo instance pixels, thus, the set of possible featues in this aea is vey lage ( featues, see [12] fo calculation details). In contast to the Haa basis functions, the set of ectangle featues is not minimal. A single featue is effectively computed on input images using integal images [25], also known as summed aea tables [12]. An integal image I is an intemediate epesentation fo the image and contains the sum of gay scale pixel values of image N with height y and width x, i.e., I(x, y) = x x =0 y =0 y N(x, y ). The integal image is computed ecusively, by the fomulas: I(x, y) = I(x, y 1) + I(x 1, y) + N(x, y) I(x 1, y 1) with I( 1, y) = I(x, 1) = I( 1, 1) = 0, theefoe equiing only one scan ove the input data. This intemediate epesentation I(x, y) allows the computation of a ectangle featue value at (x, y) with height and width (h, w) using fou efeences (see Fig. 3 (ight)): F (x, y, h, w) = I(x, y) + I(x + w, y + h) I(x, y + h) I(x + w, y). Fo computing the otated featues, Lienhat et. al. intoduced otated summed aea tables that contain the sum of the pixels of the ectangle otated by 45 with the bottom-most cone at (x, y) and extending till the boundaies of the image (see Fig. 3 (bottom left)) [12]: I (x, y) = x x =0 x x y y =0 N(x, y ). The otated integal image I is computed ecusively, i.e., I (x, y) = I (x 1, y 1) + I (x + 1, y 1) + I (x, y 1) + N(x, y) + N(x, y 1) using the stat values I ( 1, y) = I (x, 1) = I (x, 2) = I ( 1, 1) = I ( 1, 2) = 0. Fou table lookups ae equied to compute the pixel sum of any otated ectangle with the fomula: F (x, y, h, w) = I (x + w h, y + w + h 1) + I (x, y 1) I (x h, y + h 1) I (x + w, y + w 1). Since the featues ae compositions of ectangles, they ae computed with seveal lookups and subtactions weighted with the aea of the black and white ectangles. To detect a featue, a theshold is equied. This theshold is automatically detemined duing a fitting pocess, such that a minimal numbe of examples ae misclassified. Futhemoe, the etun values (α, β) of the featue ae detemined, such that the eo on the examples is minimized. The examples ae given in a set of images that ae classified as positive o negative samples. The set is also used in the leaning phase that is biefly descibed next ::::: ;;;;;; ::::: ;;;;;; ::::: ;;;;;; ::::: ;;;;;; ::::: ;;;;;; > > >???? > > >???? > > >???? > > >???? > > >???? > > AAAAA BBBBB CCCCCC BBBBB CCCCCC I (x,y 2) DDDEEEE DDDEEEE I (x,y) y GMGMGMGMGM JKJKJKJKJKJKJKJKJKJKJKJKJKJKJKJKJKJKJKJK x x+w GLGLGLGLGL GMGMGMGMGM JIJIJIJIJIJIJIJIJIJIJIJIJIJIJIJIJIJIJIJI JKJKJKJKJKJKJKJKJKJKJKJKJKJKJKJKJKJKJKJK GLGLGLGLGL GMGMGMGMGM JIJIJIJIJIJIJIJIJIJIJIJIJIJIJIJIJIJIJIJI JKJKJKJKJKJKJKJKJKJKJKJKJKJKJKJKJKJKJKJK y+h I (x h,y+h 1) w h ONONONONONONO ONONONONONONO OPOPOPOPOPOPO OPOPOPOPOPOPO POPOPOPOPOPOPOP OTOTOTOT RJRJRJRJRJRJRJRJRJRJRJRJR ONONONONO ONONONONO OPOPOPOPO OPOPOPOPO OSOSOSO OTOTOTO JQJQJQJQJQJQJQJQJQJ JRJRJRJRJRJRJRJRJRJ ONONO ONONO OPOPO OPOPO OSO OTO JQJQJQJQJQJ JRJRJRJRJRJ O JRJJ I (x h+w,y+w+h 1) I (x+w,y+w 1) Fig. 3. Top left: Edge, line, diagonal and cente suound featues ae used fo classification. Top ight: Computation of featue values F in the shaded egion is based on the fou uppe ectangles. Bottom left: Calculation of the otated integal image I. Bottom ight: Fou lookups in the otated integal image ae equied to compute the featue value of a otated featue F. (a) th. = (b) th. = th. = th. = th. = Fig. 4. Left: A simple featue classifie with its theshold and etun values α and β. Right: A Classification and Regession Tee with 4 splits. Accoding to the specific filte applied to the image input section x, the output of the tee, h(x) is calculated, depending on the theshold values. C. Leaning Classification Functions 1) Classification and Regession Tees: Fo all possible featues a Classification and Regession Tee (CART) is ceated. CART analysis is a fom of binay ecusive patitioning. Each node is split into two child nodes, the oiginal node is called a paent node. The tem ecusive efes to the fact that the binay patitioning pocess is applied ove and ove to each a given numbe of splits (i.e., 6 splits in the case of the object volksbot). In ode to find the best possible split featues, we compute all possible splits, as well as all possible etun values to be used in a split node. The pogam seeks to maximize the aveage puity of the two child nodes using the misclassification eo measue [17]. Fig. 4 (left) shows a simple featue classifie and a simple CART (ight). 2) Gentle Ada Boost fo CARTs: The Gentle Ada Boost Algoithm is a vaiant of the poweful boosting leaning technique [6]. It is used to select a set of simple CARTs to achieve a given detection and eo ate. In the following, a detection is efeed to as a hit and an eo as a false alam. The vaious Ada Boost algoithms diffe in the update scheme of the weights. Accoding to Lienhat et al., the Gentle Ada Boost Algoithm is cuently the most successful leaning pocedue tested fo face detection applications [12].
4 The leaning is based on N weighted taining examples (x 1, y 1 ),..., (x N, y N ), whee x i ae the images and y i { 1, 1} the classified output i {1,..., N}. At the beginning of the leaning phase, the weights w i ae initialized with w i = 1/N. The following thee steps ae epeated to select simple CARTs until a given detection ate d is eached: 1) Evey simple classifie, i.e., a CART, is fit to the data. Heeby the eo e is calculated with espect to the weights w i. 2) The best CART h t is chosen fo the classification function. The counte t is incemented. 3) The weights ae updated with w i := w i e yiht(xi) and enomalized. The final output of the classifie is sign( T t=1 h t(x)) > 0, with h(x) the weighted etun value of the CART. Next, a cascade based on these classifies is built. D. The Cascade of Classifies The pefomance of a single classifie is not suitable fo object classification, since it poduces a high hit ate, e.g., 0.999, but also a high eo ate, e.g., 0.5. Nevetheless, the hit ate is significantly highe than the eo ate. To constuct an oveall good classifie, seveal classifies ae aanged in a cascade, i.e., a degeneated decision tee. In evey stage of the cascade, a decision is made whethe the image contains the object o not. This computation educes both ates. Since the hit ate is close to one, thei multiplication esults also in a value close to one, while the multiplication of the smalle eo ates appoaches zeo. Futhemoe, this speeds up the whole classification pocess, since lage pats of the image do not contain elevant data. These aeas can be discaded quickly in the fist stages. An oveall effective cascade is leaned by a simple iteative method. Fo evey stage, the classification function h(x) is leaned until the equied hit ate is eached. The pocess continues with the next stage using the coectly classified positive and the cuently misclassified negative examples. These negative examples ae andom image pats geneated fom the given negative examples that pass the pevious stages and thus ae misclassified. This bootstapping pocess is the most time consuming of the taining phase. The numbe of CARTs used in each classifie may incease with additional stages. Fig. 5 shows an example cascade of classifies fo detecting a volksbot in 2D depth images, whose esults ae given in Table I. E. Application of the Cascades Seveal expeiments wee made to evaluate the pefomance of the poposed appoach with two diffeent kinds of images, namely, eflectance and depth images. Both types ae acquied by the 3D lase ange finde and ae pactically light invaiant. About 200 epesentation of the objects wee taken in addition to a wide vaiety of negative examples without any taget object. The detection stats with the smallest classifie size, e.g., pixel fo the human classifie, fo the volksbot classifie. The image is seached fom top left to bottom ight by applications of the cascade. To detect th. = th. = th. = th. = th. = th. = th. = th. = th. = th. = th. = th. = th. = th. = th. = th. = th. = th. = th. = th. = th. = th. = th. = th. = th. = th. = th. = th. = th. = th. = th. = Σ h(x) < 0 Σ h(x) < 0 th. = th. = th. = th. = th. = th. = th. = th. = 6.485e 06 th. = th. = th. = h(x) >= 0 Σ h(x) >= 0 Σ h(x) >= 0 Σ Fig. 5. The fist thee stages of a cascade of classifies to detect the object volksbot. Evey stage contains seveal simple classifie tees that use Haa-like featues with a theshold th. and etun values of P h(x). h(x) is detemined by the path though the tees. objects on lage scales, the detecto is escaled. An advantage of the Haa-like featues is that they ae easily scalable.
5 Fig. 6. Object points estimation by ay tacing. Top left: All points inside a detection aea ae extacted. Top Right and bottom: 3D view. 3D points inside the detecto aea (viewing cone) ae ed coloed. Each featue equies only a fixed numbe of look-ups in the integal image, independent of the scale. Time-consuming pictue scales ae not necessay to achieve scale invaiance. Fig. 9 show examples of the detection. To decease the false detection ate, we combine the cascades of the depth and eflectance images. Thee ae two possible ways fo combining: Eithe the two cascades un inteleaved o seial and epesent a logical and [13]. The joint cascade deceases the false detection ate close to zeo. To avoid the eduction of the hit ate, seveal diffeent offsceen endeed images ae used, whee the vitual camea is otated and the apex angle is changed [13]. A. Object Points Estimation V. OBJECT LOCALIZATION Afte object detection in a 2D pojection the algoithm finds the coesponding 3D points using ay tacing. All 3D points that have been pojected into the classified aea ae etieved using a special OpenGL pojection matix. Fig. 6 (ight) shows a endeing of aytaced 3D points. B. Model Matching Afte the 3D data (set D) that contain the object is found, a given 3D model fom the object database is matched into the point cloud. The model M is also saved as 3D point cloud in the database. The well known iteative closest points algoithm (ICP) is used to find a matching [4]. The ICP algoithm calculates iteatively the point coespondences. In each iteation step, the algoithm selects the closest points as coespondences and calculates the tansfomation, i.e., otation and tanslation (R, t) fo minimizing the equation E(R, t) = w i,j d i (Rm j + t) 2, N m N d i=1 j=1 1 N N m i (Rd i + t) 2 (1) i=1 whee N m and N d, ae the numbe of points in the model set M o data set D, espectively, and w ji ae the weights fo a point match. The weights ae assigned as follows: w ji = 1, if m i is the closest point to d j within a close limit, w ji = 0 othewise. It is shown that the iteation teminates in a minimum [4]. In each iteation, the tansfomation is calculated by the quatenion based method of Hon [9]. The assumption is that the point coespondences ae coect in the last iteation step. Finally, the pose of the model coesponds to the one in the data set. C. Evaluating the Match Geneally, one is inteested in the quality of the matching, i.e., the accuacy of the model pose inside the 3D data. Many application specific tasks equie this estimation, e.g., complicated obot navigation tasks. Howeve, the value of the eo function (1) does not give this infomation, since point densities influence this value. Diffeent point densities ae the esult of the scanning pocess, i.e., the spheical and continuous measuement of the lase. The scanne emits the lase beams in a spheical way such that the data points close to the souce ae moe dense. A competitive leaning technique is used to subsample the model and data set. 1) Competitive Object Leaning: In addition to subsampling, goals of competitive object leaning ae the minimization of the expected quantization eo and entopy maximization. A finite set of 3D scan points D is subsambled to the set A = {w 1, w 2,..., w N }. Eo minimization is done with espect to the following function: E(D, A) = 1 D w i A ξ R c ξ w i, with the set A of samples and the Voonoi set R c of unit c, i.e., R c = {ξ D s(ξ) = c} and s(ξ) = ag min c A ξ w c. Entopy maximization guaantees inheent obustness. The failue of efeence vectos, i.e., missing 3D points, fects only a limited faction of the data. Intepeting the geneation of an input signal and the subsequent mapping onto the neaest sample in A as a andom expeiment which assigns a value x A to the andom vaiable X, then maximizing the entopy H(X) = x A P (x) log(p (x)) is equivalent to equipobable samples. The following neual gas algoithm leans and subsamples 3D points clouds [7]: i.) Initialize the set A to contain N vectos, andomly fom the input set. Set t = 0.
6 _ h gh i gh i Point Distibution 50 numbe of points a ac ad ac ad ac ae ad 10 ac ae ad ac ae ad ac ae ad ae m a aq a a a 0 a a distances [cm] Fig. 8. A typical distibution of distances between closest points te egisteing two models with a fixed (hee: 250) numbe of points. Fig. 7. Top: 3D models (point clouds) of the database. Bottom: sumbsampled models with 250 points. ii.) Geneate at andom an input element ξ, i.e., select a point fom D. iii.) Ode all elements of A accoding to thei distance to ξ, i.e., find the sequence of indices (i 0, i 1,..., i N 1 ) such that w i0 is the efeence vecto closest to ξ, w i1 is the efeence vecto second closest to ξ, etc., w ik, k = 0,..., N 1 is the efeence vecto such that k vectos w j exists that ae close to ξ than w ik. k i (ξ, A) denotes the numbe k associated with w i. iv.) Adapt the efeence vectos accoding to w i = ε(t)h λ (k i (ξ, A)) (ξ w i ), with the following time dependencies: λ(t) = λ i (λ f /λ i ) t/tmax, ε(t) = ε i (ε f /ε i ) t/tmax, h λ (k) = exp( k/λ(t)). v.) Incease the time paamete t. The neual gas algoithms is used with the following paametes: λ f = 0.01, λ i = 10.0, ε i = 0.5, ε f = 0.005, t max = Note that t max contols the un time. Fig. 7 shows 3D models of the database (top ow) and subsampled vesions (bottom) with 250 points. 2) Estimating Matching Quality: Given two egisteed point sets that contain an equal numbe of points, e.g., 250 points deived unde the pemise of minimization of the expected quantization eo and entopy maximization, the quality of a matching can be evaluated using the following method: The distibution of shotest distances d ij between the ith and the jth point (closest points) te egisteing two models with a fixed (hee: 250) numbe of points show a typical stuctue (Fig. 8). Many distances ae vey small, i.e., less than 0.3 cm, and thee ae also many lage distances, e.g., geate than 1 cm. To ou expeience it is always easy to find a good theshold to sepaate the two maximas. Afte dividing the set of distances d i, the algoithm computes the mean and the standad deviation of the matching, i.e., µ = 1 N N d i σ = 1 N N (d i µ) 2 i=1 i=1 Based on these values one estimates the matching quality by computing a measue D as a function of µ and σ (we have been using D = µ + 3σ). Small values of D coespond to a high quality matching wheeas inceasing values epesent lowe qualities. VI. RESULTS AND CONCLUSION The pocess of geneating the cascade of classifies is elatively time-consuming, but it poduces quite pomising esults. The fist thee stages of a leaned cascade ae shown in Fig. 5. The time pefomance of the object detection cucially depends on the bootstapping, i.e., on the geneation of false positive examples duing the stage leaning. Nevetheless, leaning has to be executed only once, the application of the cascade if vey fast (300 ms). Thus the majo time fo the accuate object localization is spent duing the model alignment and evaluation step ( 1.4 s). The capabilities of the chosen appoach have been evaluated in vaious expeiments. Fig. 9 shows fou examples of successful detections and Table I summaizes the object localization esults. A. Futue Wok Needless to say, much wok emains to be done. Futue wok will concentate on fou majo aspects: 1) Impove the computational pefomance of the system to impove obot/envionment inteaction. 2) Geneate high level desciptions and semantic maps including the 3D infomation, e.g., in XML fomat.
7 Fig. 9. Examples of object detection and localization. Fom Left to ight: (1) Detection using single cascade of classifies. Geen: detection in eflection image, yellow: detection in depth image. (2) Detection using the combined cascade. (3) Supeimposed to the depth image is the matched 3D model. (4) Detected object in the aw scanne data, i.e., point epesentation. TABLE I OBJECT NAME, NUMBER OF STAGES USED FOR CLASSIFICATION VERSUS HIT RATE AND THE TOTAL NUMBER OF FALSE ALARMS USING THE SINGLE AND COMBINED CASCADES. THE TEST SETS CONSIST OF 89 IMAGES RENDERED FROM 20 3D SCANS. THE AVERAGE PROCESSING TIME IS ALSO GIVEN, INCLUDING THE RENDERING, CLASSIFICATION, RAY TRACING, MATCHING AND EVALUATION TIME. object # stages detection ate (eflect. img. / depth img.) false alams (eflect. img. / depth img.) aveage poc. time chai (0.867 / 0.767) 12 (47 / 33) 1.9 sec kut obot (0.912 / 0.947) 0 ( 5 / 7) 1.7 sec volksbot obot (0.844 / 0.851) 5 (42 / 23) 2.3 sec human (0.963 / 0.961) 1 (13 / 17) 1.6 sec
8 The semantic maps will contain spatial 3D data with desciptions and labels. 3) Integate a camea and enhance the semantic intepetation by fusing colo images with ange data. The apetue angle of the camea will be enlaged using a pan and tilt unit to acquie colo infomation fo all measued ange points. 4) Enlage the database with moe objects of an indoo and outdoo envionment and build an explicit knowledge base, i.e., specifying a semantic net containing geneal object elations as well as links to the object database [14]. The final goal of object detection and localization is to develop unesticted, automatic and highly eliable algoithms that could be used in scenaios like RoboCup Rescue. B. Conclusions This pape has pesented a novel method fo the leaning, fast detection and localization of instances of 3D object classes. The 3D ange and eflectance lase scanne data ae tansfomed into images by off-sceen endeing. Fo fast object detection, a cascade of classifies is built, i.e., a linea decision tee [25]. The classifies ae composed of classification and egession tees (CARTs) and model the objects with thei view dependencies. Each CART makes its decisions based on featue classifies. The featues ae edge, line, cente suound, o otated featues. Afte object detection the object is localized using a point matching stategy. The pose is detemined with six degees of feedom, i.e., with espect to the x, y, and z positions and the oll, yaw and pitch angles. A final computation etuns a quality measue fo the object localization. The pesented combination of algoithms, i.e., the system achitectue enables high accuate, fast and eliable 3D object localization fo autonomous mobile obots. REFERENCES [1] The volksbot obot, volksbot/, [2] P. Allen, I. Stamos, A. Gueoguiev, E. Gold, and P. Blae. AVENUE: Automated Site Modeling in Uban Envionments. In Poceedings of the thid Intenational Confeence on 3D Digital Imaging and Modeling (3DIM 01), Quebec City, Canada, May [3] A. P. Ashock, R. B. Fishe, C. Robetson, and N. Weghi. Finding suface coespondences fo object ecognition and egistation using paiwise histoams. In Poceedings of the Euopean Confeence on Compute Vision (ECCV 98), pages , Feibug, Gemany, June [4] P. Besl and N. McKay. A method fo Registation of 3 D Shapes. IEEE Tansactions on Patten Analysis and Machine Intelligence, 14(2): , Febuay [5] F. Dellaet, W. Bugad, D. Fox, and S. Thun. Using the Condensation Algoithm fo Robust, Vision-based Mobile Robot Localization. In Poceedings of the IEEE Confeence on Compute Vision and Patten Recognition (CVPR 99), Ft. Collins, USA, June [6] Y. Feund and R. E. Schapie. Expeiments with a new boosting algoithm. In Machine Leaning: Poceedings of the 13th Intenational Confeence, pages , [7] B. Fitzke. A gowing neual gas netwok leans topologies. In Advances in Neual Infomation Pocessing Systems 7 - Poceedings of the 7th Advances in Neual Infomation Pocessing Systems (NIPS 95), pages , Cambidge, MA, USA, [8] A. Haa. Zu Theoie de othogonalen Funktionensysteme. Mathematische Annalen, (69): , [9] B. Hon. Closed fom solution of absolute oientation using unit quatenions. Jounal of the Optical Society of Ameica A, 4(4): , Apil [10] A. Johnson and M. Hebet. Using spin images fo efficient object ecognition in clutteed 3D scenes. IEEE Tansactions on Patten Analysis and Machine Intelligence, 21(5): , May [11] F. Launay, A. Ohya, and S. Yuta. Autonomous Indoo Mobile Robot Navigation by detecting Fluoescent Tubes. In Poccedings of the 10th Intenational Confeence on Advanced Robotics (ICAR 01), Budapest, Hungay, August [12] R. Lienhat and J. Maydt. An Extended Set of Haa-like Featues fo Rapid Object Detection. In Poceedings of the IEEE Confeence on Image Pocessing (ICIP 02), pages , New Yok, USA, Septmbe [13] A. Nüchte, H. Sumann,, and J. Hetzbeg. Automatic Classification of Objects in 3D Lase Range Scans. In Poceedings of the 8th Confeence on Intelligent Autonomous Systems (IAS 04), pages , Amstedam, The Nethelands, Mach [14] A. Nüchte, H. Sumann, and J. Hetzbeg. Automatic Model Refinement fo 3D Reconstuction with Mobile Robots. In Poceedings of the 4th IEEE Intenational Confeence on Recent Advances in 3D Digital Imaging and Modeling (3DIM 03), pages , Banff, Canada, Octobe [15] C. Papageogiou, M. Oen, and T. Poggio. A geneal famewok fo object detection. In Poceedings of the 6th Intenational Confeence on Compute Vision (ICCV 98), Bombay, India, Januay [16] S. Ruiz-Coea, L. G. Shapio, and M. Meila. A New Paadigm fo Recognizing 3-D Object Shapes fom Range Data. In Poceedings of the IEEE Confeence on Compute Vision and Patten Recognition (CVPR 03), Madison, USA, June [17] S. Russell and P. Novig. Atificial Intelligence, A Moden Appoach. Pentice Hall, Inc., Uppe Sanddle Rive, NJ, USA, [18] S. Se, D. Lowe, and J. Little. Local and Global Localization fo Mobile Robots using Visual Landmaks. In Poceedings of the IEEE/RSJ Intenational Confeence on Intelligent Robots and Systems (IROS 01), Hawaii, USA, Octobe [19] V. Sequeia, K. Ng, E. Wolfat, J. Goncalves, and D. Hogg. Automated 3D econstuction of inteios with multiple scan views. In Poceedings of SPIE, Electonic Imaging 99, The Society fo Imaging Science and Technology /SPIE s 11th Annual Symposium, San Jose, CA, USA, Januay [20] F. Stein and G. Medioni. Stuctual indexing: Efficient 3d object ecognition. Tansaction on Patten Analysis and machine Vision (PAMI), 14: , Febuay [21] Y. Sun, J. Paik, A. Koschan, D. Page, and M. Abidi. Point Fingepint: An New 3D Object Repesention Scheme. IEEE tansaction on Systems, Man, and Cybenetics Pat B: Cybenetics, 33(4), [22] H. Sumann, K. Lingemann, A. Nüchte, and J. Hetzbeg. A 3D lase ange finde fo autonomous mobile obots. In Poceedings of the of the 32nd Intenational Symposium on Robotics (ISR 01), pages , Seoul, Koea, Apil [23] H. Sumann, A. Nüchte, and J. Hetzbeg. An autonomous mobile obot with a 3D lase ange finde fo 3D exploation and digitalization of indoo en vionments. Robotics and Autonomous Systems, 45(3 4): , Decembe [24] S. Thun, D. Fox, and W. Bugad. A eal-time algoithm fo mobile obot mapping with application to multi obot and 3D mapping. In Poceedings of the IEEE Intenational Confeence on Robotics and Automation (ICRA 00), San Fancisco, CA, USA, Apil [25] Paul Viola and Michael J. Jones. Robust eal-time face detection. Intenational Jounal of Compute Vision, 57(2): , May [26] D. Zhang and M. Hebet. Hamonic maps and thei application in suface matching. In Poceedings of the IEEE Confeence on Compute Vision and Patten Recognition (CVPR 99), pages , Ft. Collins, CO, USA, June ACKNOWLEDGMENTS The wok was done duing the authos time at the Faunhofe Institute fo Autonomous intelligent Systems. We would like to thank Saa Miti, Simone Fintop, Kai Pevölz and Matthias Hennig.
Robust Object Detection at Regions of Interest with an Application in Ball Recognition
Robust Object Detection at Regions of Inteest with an Application in Ball Recognition Saa Miti, Simone Fintop, Kai Pevölz, Hatmut Sumann Faunhofe Institute fo Autonomous Intelligent Systems (AIS) Schloss
More informationPositioning of a robot based on binocular vision for hand / foot fusion Long Han
2nd Intenational Confeence on Advances in Mechanical Engineeing and Industial Infomatics (AMEII 26) Positioning of a obot based on binocula vision fo hand / foot fusion Long Han Compute Science and Technology,
More informationRobust Object Detection at Regions of Interest with an Application in Ball Recognition
Robust Object Detection at Regions of Inteest with an Application in Ball Recognition Saa Miti, Simone Fintop, Kai Pevölz, Hatmut Sumann Faunhofe Institute fo Autonomous Intelligent Systems (AIS) Schloss
More informationJournal of World s Electrical Engineering and Technology J. World. Elect. Eng. Tech. 1(1): 12-16, 2012
2011, Scienceline Publication www.science-line.com Jounal of Wold s Electical Engineeing and Technology J. Wold. Elect. Eng. Tech. 1(1): 12-16, 2012 JWEET An Efficient Algoithm fo Lip Segmentation in Colo
More informationDetection and Recognition of Alert Traffic Signs
Detection and Recognition of Alet Taffic Signs Chia-Hsiung Chen, Macus Chen, and Tianshi Gao 1 Stanfod Univesity Stanfod, CA 9305 {echchen, macuscc, tianshig}@stanfod.edu Abstact Taffic signs povide dives
More informationIllumination methods for optical wear detection
Illumination methods fo optical wea detection 1 J. Zhang, 2 P.P.L.Regtien 1 VIMEC Applied Vision Technology, Coy 43, 5653 LC Eindhoven, The Nethelands Email: jianbo.zhang@gmail.com 2 Faculty Electical
More informationSegmentation of Casting Defects in X-Ray Images Based on Fractal Dimension
17th Wold Confeence on Nondestuctive Testing, 25-28 Oct 2008, Shanghai, China Segmentation of Casting Defects in X-Ray Images Based on Factal Dimension Jue WANG 1, Xiaoqin HOU 2, Yufang CAI 3 ICT Reseach
More informationAn Unsupervised Segmentation Framework For Texture Image Queries
An Unsupevised Segmentation Famewok Fo Textue Image Queies Shu-Ching Chen Distibuted Multimedia Infomation System Laboatoy School of Compute Science Floida Intenational Univesity Miami, FL 33199, USA chens@cs.fiu.edu
More informationIP Network Design by Modified Branch Exchange Method
Received: June 7, 207 98 IP Netwok Design by Modified Banch Method Kaiat Jaoenat Natchamol Sichumoenattana 2* Faculty of Engineeing at Kamphaeng Saen, Kasetsat Univesity, Thailand 2 Faculty of Management
More informationControlled Information Maximization for SOM Knowledge Induced Learning
3 Int'l Conf. Atificial Intelligence ICAI'5 Contolled Infomation Maximization fo SOM Knowledge Induced Leaning Ryotao Kamimua IT Education Cente and Gaduate School of Science and Technology, Tokai Univeisity
More informationColor Correction Using 3D Multiview Geometry
Colo Coection Using 3D Multiview Geomety Dong-Won Shin and Yo-Sung Ho Gwangju Institute of Science and Technology (GIST) 13 Cheomdan-gwagio, Buk-ku, Gwangju 500-71, Republic of Koea ABSTRACT Recently,
More informationOptical Flow for Large Motion Using Gradient Technique
SERBIAN JOURNAL OF ELECTRICAL ENGINEERING Vol. 3, No. 1, June 2006, 103-113 Optical Flow fo Lage Motion Using Gadient Technique Md. Moshaof Hossain Sake 1, Kamal Bechkoum 2, K.K. Islam 1 Abstact: In this
More informationObstacle Avoidance of Autonomous Mobile Robot using Stereo Vision Sensor
Obstacle Avoidance of Autonomous Mobile Robot using Steeo Vision Senso Masako Kumano Akihisa Ohya Shin ichi Yuta Intelligent Robot Laboatoy Univesity of Tsukuba, Ibaaki, 35-8573 Japan E-mail: {masako,
More informationA modal estimation based multitype sensor placement method
A modal estimation based multitype senso placement method *Xue-Yang Pei 1), Ting-Hua Yi 2) and Hong-Nan Li 3) 1),)2),3) School of Civil Engineeing, Dalian Univesity of Technology, Dalian 116023, China;
More informationA Two-stage and Parameter-free Binarization Method for Degraded Document Images
A Two-stage and Paamete-fee Binaization Method fo Degaded Document Images Yung-Hsiang Chiu 1, Kuo-Liang Chung 1, Yong-Huai Huang 2, Wei-Ning Yang 3, Chi-Huang Liao 4 1 Depatment of Compute Science and
More informationImproved Fourier-transform profilometry
Impoved Fouie-tansfom pofilomety Xianfu Mao, Wenjing Chen, and Xianyu Su An impoved optical geomety of the pojected-finge pofilomety technique, in which the exit pupil of the pojecting lens and the entance
More informationProf. Feng Liu. Fall /17/2016
Pof. Feng Liu Fall 26 http://www.cs.pdx.edu/~fliu/couses/cs447/ /7/26 Last time Compositing NPR 3D Gaphics Toolkits Tansfomations 2 Today 3D Tansfomations The Viewing Pipeline Mid-tem: in class, Nov. 2
More informationTopic -3 Image Enhancement
Topic -3 Image Enhancement (Pat 1) DIP: Details Digital Image Pocessing Digital Image Chaacteistics Spatial Spectal Gay-level Histogam DFT DCT Pe-Pocessing Enhancement Restoation Point Pocessing Masking
More informationAugmented Reality. Integrating Computer Graphics with Computer Vision Mihran Tuceryan. August 16, 1998 ICPR 98 1
Augmented Reality Integating Compute Gaphics with Compute Vision Mihan Tuceyan August 6, 998 ICPR 98 Definition XCombines eal and vitual wolds and objects XIt is inteactive and eal-time XThe inteaction
More informationImage Enhancement in the Spatial Domain. Spatial Domain
8-- Spatial Domain Image Enhancement in the Spatial Domain What is spatial domain The space whee all pixels fom an image In spatial domain we can epesent an image by f( whee x and y ae coodinates along
More informationSpiral Recognition Methodology and Its Application for Recognition of Chinese Bank Checks
Spial Recognition Methodology and Its Application fo Recognition of Chinese Bank Checks Hanshen Tang 1, Emmanuel Augustin 2, Ching Y. Suen 1, Olivie Baet 2, Mohamed Cheiet 3 1 Cente fo Patten Recognition
More informationModule 6 STILL IMAGE COMPRESSION STANDARDS
Module 6 STILL IMAE COMPRESSION STANDARDS Lesson 17 JPE-2000 Achitectue and Featues Instuctional Objectives At the end of this lesson, the students should be able to: 1. State the shotcomings of JPE standad.
More information(a, b) x y r. For this problem, is a point in the - coordinate plane and is a positive number.
Illustative G-C Simila cicles Alignments to Content Standads: G-C.A. Task (a, b) x y Fo this poblem, is a point in the - coodinate plane and is a positive numbe. a. Using a tanslation and a dilation, show
More informationA Novel Automatic White Balance Method For Digital Still Cameras
A Novel Automatic White Balance Method Fo Digital Still Cameas Ching-Chih Weng 1, Home Chen 1,2, and Chiou-Shann Fuh 3 Depatment of Electical Engineeing, 2 3 Gaduate Institute of Communication Engineeing
More informationA New and Efficient 2D Collision Detection Method Based on Contact Theory Xiaolong CHENG, Jun XIAO a, Ying WANG, Qinghai MIAO, Jian XUE
5th Intenational Confeence on Advanced Mateials and Compute Science (ICAMCS 2016) A New and Efficient 2D Collision Detection Method Based on Contact Theoy Xiaolong CHENG, Jun XIAO a, Ying WANG, Qinghai
More informationFifth Wheel Modelling and Testing
Fifth heel Modelling and Testing en Masoy Mechanical Engineeing Depatment Floida Atlantic Univesity Boca aton, FL 4 Lois Malaptias IFMA Institut Fancais De Mechanique Advancee ampus De lemont Feand Les
More informationLecture # 04. Image Enhancement in Spatial Domain
Digital Image Pocessing CP-7008 Lectue # 04 Image Enhancement in Spatial Domain Fall 2011 2 domains Spatial Domain : (image plane) Techniques ae based on diect manipulation of pixels in an image Fequency
More informationAn Extension to the Local Binary Patterns for Image Retrieval
, pp.81-85 http://x.oi.og/10.14257/astl.2014.45.16 An Extension to the Local Binay Pattens fo Image Retieval Zhize Wu, Yu Xia, Shouhong Wan School of Compute Science an Technology, Univesity of Science
More informationEYE DIRECTION BY STEREO IMAGE PROCESSING USING CORNEAL REFLECTION ON AN IRIS
EYE DIRECTION BY STEREO IMAGE PROCESSING USING CORNEAL REFLECTION ON AN IRIS Kumiko Tsuji Fukuoka Medical technology Teikyo Univesity 4-3-14 Shin-Katsutachi-Machi Ohmuta Fukuoka 836 Japan email: c746g@wisdomcckyushu-uacjp
More informationMULTI-TEMPORAL AND MULTI-SENSOR IMAGE MATCHING BASED ON LOCAL FREQUENCY INFORMATION
Intenational Achives of the Photogammety Remote Sensing and Spatial Infomation Sciences Volume XXXIX-B3 2012 XXII ISPRS Congess 25 August 01 Septembe 2012 Melboune Austalia MULTI-TEMPORAL AND MULTI-SENSOR
More informationINDEXATION OF WEB PAGES BASED ON THEIR VISUAL RENDERING
INDEXATION OF WEB PAGES BASED ON THEIR VISUAL RENDERING Emmanuel Buno Univesité du Sud Toulon-Va / LSIS CNRS BP 20132, F-83957 La Gade buno@univ-tln.f Nicolas Faessel LSIS CNRS Domaine Univesitaie de Saint-Jéôme
More informationADDING REALISM TO SOURCE CHARACTERIZATION USING A GENETIC ALGORITHM
ADDING REALISM TO SOURCE CHARACTERIZATION USING A GENETIC ALGORITHM Luna M. Rodiguez*, Sue Ellen Haupt, and Geoge S. Young Depatment of Meteoology and Applied Reseach Laboatoy The Pennsylvania State Univesity,
More informationMono Vision Based Construction of Elevation Maps in Indoor Environments
8th WSEAS Intenational onfeence on SIGNAL PROESSING, OMPUTATIONAL GEOMETRY and ARTIFIIAL VISION (ISGAV 08) Rhodes, Geece, August 0-, 008 Mono Vision Based onstuction of Elevation Maps in Indoo Envionments
More informationSeveral algorithms exist to extract edges from point. system. the line is computed using a least squares method.
Fast Mapping using the Log-Hough Tansfomation B. Giesle, R. Gaf, R. Dillmann Institute fo Pocess Contol and Robotics (IPR) Univesity of Kalsuhe D-7618 Kalsuhe, Gemany fgieslejgafjdillmanng@ia.uka.de C.F.R.
More informationCSE 165: 3D User Interaction
CSE 165: 3D Use Inteaction Lectue #6: Selection Instucto: Jugen Schulze, Ph.D. 2 Announcements Homewok Assignment #2 Due Fiday, Januay 23 d at 1:00pm 3 4 Selection and Manipulation 5 Why ae Selection and
More information3D inspection system for manufactured machine parts
3D inspection system fo manufactued machine pats D. Gacía a*, J. M. Sebastián a*, F. M. Sánchez a*, L. M. Jiménez b*, J. M. González a* a Dept. of System Engineeing and Automatic Contol. Polytechnic Univesity
More informationHISTOGRAMS are an important statistic reflecting the
JOURNAL OF L A T E X CLASS FILES, VOL. 14, NO. 8, AUGUST 2015 1 D 2 HistoSketch: Disciminative and Dynamic Similaity-Peseving Sketching of Steaming Histogams Dingqi Yang, Bin Li, Laua Rettig, and Philippe
More information3D Reconstruction from 360 x 360 Mosaics 1
CENTER FOR MACHINE PERCEPTION 3D Reconstuction fom 36 x 36 Mosaics CZECH TECHNICAL UNIVERSITY {bakstein, pajdla}@cmp.felk.cvut.cz REPRINT Hynek Bakstein and Tomáš Pajdla, 3D Reconstuction fom 36 x 36 Mosaics,
More informationTowards Adaptive Information Merging Using Selected XML Fragments
Towads Adaptive Infomation Meging Using Selected XML Fagments Ho-Lam Lau and Wilfed Ng Depatment of Compute Science and Engineeing, The Hong Kong Univesity of Science and Technology, Hong Kong {lauhl,
More informationImage Registration among UAV Image Sequence and Google Satellite Image Under Quality Mismatch
0 th Intenational Confeence on ITS Telecommunications Image Registation among UAV Image Sequence and Google Satellite Image Unde Quality Mismatch Shih-Ming Huang and Ching-Chun Huang Depatment of Electical
More informationA Minutiae-based Fingerprint Matching Algorithm Using Phase Correlation
A Minutiae-based Fingepint Matching Algoithm Using Phase Coelation Autho Chen, Weiping, Gao, Yongsheng Published 2007 Confeence Title Digital Image Computing: Techniques and Applications DOI https://doi.og/10.1109/dicta.2007.4426801
More informationView Synthesis using Depth Map for 3D Video
View Synthesis using Depth Map fo 3D Video Cheon Lee and Yo-Sung Ho Gwangju Institute of Science and Technology (GIST) 1 Oyong-dong, Buk-gu, Gwangju, 500-712, Republic of Koea E-mail: {leecheon, hoyo}@gist.ac.k
More informationRANDOM IRREGULAR BLOCK-HIERARCHICAL NETWORKS: ALGORITHMS FOR COMPUTATION OF MAIN PROPERTIES
RANDOM IRREGULAR BLOCK-HIERARCHICAL NETWORKS: ALGORITHMS FOR COMPUTATION OF MAIN PROPERTIES Svetlana Avetisyan Mikayel Samvelyan* Matun Kaapetyan Yeevan State Univesity Abstact In this pape, the class
More informationA Memory Efficient Array Architecture for Real-Time Motion Estimation
A Memoy Efficient Aay Achitectue fo Real-Time Motion Estimation Vasily G. Moshnyaga and Keikichi Tamau Depatment of Electonics & Communication, Kyoto Univesity Sakyo-ku, Yoshida-Honmachi, Kyoto 66-1, JAPAN
More informationPoint-Biserial Correlation Analysis of Fuzzy Attributes
Appl Math Inf Sci 6 No S pp 439S-444S (0 Applied Mathematics & Infomation Sciences An Intenational Jounal @ 0 NSP Natual Sciences Publishing o Point-iseial oelation Analysis of Fuzzy Attibutes Hao-En hueh
More informationKalman filter correction with rational non-linear functions: Application to Visual-SLAM
1 Kalman filte coection with ational non-linea functions: Application to Visual-SLAM Thomas Féaud, Roland Chapuis, Romuald Aufèe and Paul Checchin Clemont Univesité, Univesité Blaise Pascal, LASMEA UMR
More informationMapReduce Optimizations and Algorithms 2015 Professor Sasu Tarkoma
apreduce Optimizations and Algoithms 2015 Pofesso Sasu Takoma www.cs.helsinki.fi Optimizations Reduce tasks cannot stat befoe the whole map phase is complete Thus single slow machine can slow down the
More informationDevelopment and Analysis of a Real-Time Human Motion Tracking System
Development and Analysis of a Real-Time Human Motion Tacking System Jason P. Luck 1,2 Chistian Debunne 1 William Hoff 1 Qiang He 1 Daniel E. Small 2 1 Coloado School of Mines 2 Sandia National Labs Engineeing
More informationCardiac C-Arm CT. SNR Enhancement by Combining Multiple Retrospectively Motion Corrected FDK-Like Reconstructions
Cadiac C-Am CT SNR Enhancement by Combining Multiple Retospectively Motion Coected FDK-Like Reconstuctions M. Pümme 1, L. Wigstöm 2,3, R. Fahig 2, G. Lauitsch 4, J. Honegge 1 1 Institute of Patten Recognition,
More informationStructured Light Stereoscopic Imaging with Dynamic Pseudo-random Patterns
Stuctued Light Steeoscopic Imaging with Dynamic Pseudo-andom Pattens Piee Payeu and Danick Desjadins Univesity of Ottawa, SITE, 800 King Edwad, Ottawa, ON, Canada, K1N 6N5 {ppayeu,ddesjad}@site.uottawa.ca
More informationIP Multicast Simulation in OPNET
IP Multicast Simulation in OPNET Xin Wang, Chien-Ming Yu, Henning Schulzinne Paul A. Stipe Columbia Univesity Reutes Depatment of Compute Science 88 Pakway Dive South New Yok, New Yok Hauppuage, New Yok
More informationA Mathematical Implementation of a Global Human Walking Model with Real-Time Kinematic Personification by Boulic, Thalmann and Thalmann.
A Mathematical Implementation of a Global Human Walking Model with Real-Time Kinematic Pesonification by Boulic, Thalmann and Thalmann. Mashall Badley National Cente fo Physical Acoustics Univesity of
More information10/29/2010. Rendering techniques. Global Illumination. Local Illumination methods. Today : Global Illumination Modules and Methods
Rendeing techniques Compute Gaphics Lectue 10 Can be classified as Local Illumination techniques Global Illumination techniques Global Illumination 1: Ray Tacing and Radiosity Taku Komua 1 Local Illumination
More informationGoal. Rendering Complex Scenes on Mobile Terminals or on the web. Rendering on Mobile Terminals. Rendering on Mobile Terminals. Walking through images
Goal Walking though s -------------------------------------------- Kadi Bouatouch IRISA Univesité de Rennes I, Fance Rendeing Comple Scenes on Mobile Teminals o on the web Rendeing on Mobile Teminals Rendeing
More informationFrequency Domain Approach for Face Recognition Using Optical Vanderlugt Filters
Optics and Photonics Jounal, 016, 6, 94-100 Published Online August 016 in SciRes. http://www.scip.og/jounal/opj http://dx.doi.og/10.436/opj.016.68b016 Fequency Domain Appoach fo Face Recognition Using
More informationCOMPARISON OF CHIRP SCALING AND WAVENUMBER DOMAIN ALGORITHMS FOR AIRBORNE LOW FREQUENCY SAR DATA PROCESSING
COMPARISON OF CHIRP SCALING AND WAVENUMBER DOMAIN ALGORITHMS FOR AIRBORNE LOW FREQUENCY SAR DATA PROCESSING A. Potsis a, A. Reigbe b, E. Alivisatos a, A. Moeia c,and N. Uzunoglu a a National Technical
More information4.2. Co-terminal and Related Angles. Investigate
.2 Co-teminal and Related Angles Tigonometic atios can be used to model quantities such as
More informationTHE SOLID IMAGE: a new concept and its applications
THE SOLID IMAGE: a new concept and its applications Leando Bonaz ( # ), Segio Dequal ( # ) ( # ) Politecnico di Toino - Dipatimento di Geoisose e Teitoio C.so Duca degli Abuzzi, 4 119 Toino Tel. +39.11.564.7687
More informationAll lengths in meters. E = = 7800 kg/m 3
Poblem desciption In this poblem, we apply the component mode synthesis (CMS) technique to a simple beam model. 2 0.02 0.02 All lengths in metes. E = 2.07 10 11 N/m 2 = 7800 kg/m 3 The beam is a fee-fee
More informationA ROI Focusing Mechanism for Digital Cameras
A ROI Focusing Mechanism fo Digital Cameas Chu-Hui Lee, Meng-Feng Lin, Chun-Ming Huang, and Chun-Wei Hsu Abstact With the development and application of digital technologies, the digital camea is moe popula
More information17/5/2009. Introduction
7/5/9 Steeo Imaging Intoduction Eample of Human Vision Peception of Depth fom Left and ight eye images Diffeence in elative position of object in left and ight eyes. Depth infomation in the views?? 7/5/9
More informationShortest Paths for a Two-Robot Rendez-Vous
Shotest Paths fo a Two-Robot Rendez-Vous Eik L Wyntes Joseph S B Mitchell y Abstact In this pape, we conside an optimal motion planning poblem fo a pai of point obots in a plana envionment with polygonal
More informationColor Interpolation for Single CCD Color Camera
Colo Intepolation fo Single CCD Colo Camea Yi-Ming Wu, Chiou-Shann Fuh, and Jui-Pin Hsu Depatment of Compute Science and Infomation Engineeing, National Taian Univesit, Taipei, Taian Email: 88036@csie.ntu.edu.t;
More informationInput Layer f = 2 f = 0 f = f = 3 1,16 1,1 1,2 1,3 2, ,2 3,3 3,16. f = 1. f = Output Layer
Using the Gow-And-Pune Netwok to Solve Poblems of Lage Dimensionality B.J. Biedis and T.D. Gedeon School of Compute Science & Engineeing The Univesity of New South Wales Sydney NSW 2052 AUSTRALIA bbiedis@cse.unsw.edu.au
More informationA Recommender System for Online Personalization in the WUM Applications
A Recommende System fo Online Pesonalization in the WUM Applications Mehdad Jalali 1, Nowati Mustapha 2, Ali Mamat 2, Md. Nasi B Sulaiman 2 Abstact foeseeing of use futue movements and intentions based
More informationTopological Characteristic of Wireless Network
Topological Chaacteistic of Wieless Netwok Its Application to Node Placement Algoithm Husnu Sane Naman 1 Outline Backgound Motivation Papes and Contibutions Fist Pape Second Pape Thid Pape Futue Woks Refeences
More informationTitle. Author(s)NOMURA, K.; MOROOKA, S. Issue Date Doc URL. Type. Note. File Information
Title CALCULATION FORMULA FOR A MAXIMUM BENDING MOMENT AND THE TRIANGULAR SLAB WITH CONSIDERING EFFECT OF SUPPO UNIFORM LOAD Autho(s)NOMURA, K.; MOROOKA, S. Issue Date 2013-09-11 Doc URL http://hdl.handle.net/2115/54220
More informationA Neural Network Model for Storing and Retrieving 2D Images of Rotated 3D Object Using Principal Components
A Neual Netwok Model fo Stong and Reteving 2D Images of Rotated 3D Object Using Pncipal Components Tsukasa AMANO, Shuichi KUROGI, Ayako EGUCHI, Takeshi NISHIDA, Yasuhio FUCHIKAWA Depatment of Contol Engineeng,
More informationEffects of Model Complexity on Generalization Performance of Convolutional Neural Networks
Effects of Model Complexity on Genealization Pefomance of Convolutional Neual Netwoks Tae-Jun Kim 1, Dongsu Zhang 2, and Joon Shik Kim 3 1 Seoul National Univesity, Seoul 151-742, Koea, E-mail: tjkim@bi.snu.ac.k
More informationa Not yet implemented in current version SPARK: Research Kit Pointer Analysis Parameters Soot Pointer analysis. Objectives
SPARK: Soot Reseach Kit Ondřej Lhoták Objectives Spak is a modula toolkit fo flow-insensitive may points-to analyses fo Java, which enables expeimentation with: vaious paametes of pointe analyses which
More informationLecture 27: Voronoi Diagrams
We say that two points u, v Y ae in the same connected component of Y if thee is a path in R N fom u to v such that all the points along the path ae in the set Y. (Thee ae two connected components in the
More informationTransmission Lines Modeling Based on Vector Fitting Algorithm and RLC Active/Passive Filter Design
Tansmission Lines Modeling Based on Vecto Fitting Algoithm and RLC Active/Passive Filte Design Ahmed Qasim Tuki a,*, Nashien Fazilah Mailah b, Mohammad Lutfi Othman c, Ahmad H. Saby d Cente fo Advanced
More informationCellular Neural Network Based PTV
3th Int Symp on Applications of Lase Techniques to Fluid Mechanics Lisbon, Potugal, 6-9 June, 006 Cellula Neual Netwok Based PT Kazuo Ohmi, Achyut Sapkota : Depatment of Infomation Systems Engineeing,
More informationShape Matching / Object Recognition
Image Pocessing - Lesson 4 Poduction Line object classification Object Recognition Shape Repesentation Coelation Methods Nomalized Coelation Local Methods Featue Matching Coespondence Poblem Alignment
More informationVoting-Based Grouping and Interpretation of Visual Motion
Voting-Based Gouping and Intepetation of Visual Motion Micea Nicolescu Depatment of Compute Science Univesity of Nevada, Reno Reno, NV 89557 micea@cs.un.edu Géad Medioni Integated Media Systems Cente Univesity
More information9-2. Camera Calibration Method for Far Range Stereovision Sensors Used in Vehicles. Tiberiu Marita, Florin Oniga, Sergiu Nedevschi
9-2 Camea Calibation Method fo Fa Range Steeovision Sensos Used in Vehicles ibeiu Maita, Floin Oniga, Segiu Nedevschi Compute Science Depatment echnical Univesity of Cluj-Napoca Cluj-Napoca, 400020, ROMNI
More informationAn Optimised Density Based Clustering Algorithm
Intenational Jounal of Compute Applications (0975 8887) Volume 6 No.9, Septembe 010 An Optimised Density Based Clusteing Algoithm J. Hencil Pete Depatment of Compute Science St. Xavie s College, Palayamkottai,
More informationFast quality-guided flood-fill phase unwrapping algorithm for three-dimensional fringe pattern profilometry
Univesity of Wollongong Reseach Online Faculty of Infomatics - Papes (Achive) Faculty of Engineeing and Infomation Sciences 2010 Fast quality-guided flood-fill phase unwapping algoithm fo thee-dimensional
More informationMonte Carlo Techniques for Rendering
Monte Calo Techniques fo Rendeing CS 517 Fall 2002 Compute Science Conell Univesity Announcements No ectue on Thusday Instead, attend Steven Gotle, Havad Upson Hall B17, 4:15-5:15 (efeshments ealie) Geomety
More informationAdaptation of Motion Capture Data of Human Arms to a Humanoid Robot Using Optimization
ICCAS25 June 2-5, KINTEX, Gyeonggi-Do, Koea Adaptation of Motion Captue Data of Human Ams to a Humanoid Robot Using Optimization ChangHwan Kim and Doik Kim Intelligent Robotics Reseach Cente, Koea Institute
More informationVisual Servoing from Deep Neural Networks
Visual Sevoing fom Deep Neual Netwoks Quentin Bateux 1, Eic Machand 1, Jügen Leitne 2, Fançois Chaumette 3, Pete Coke 2 Abstact We pesent a deep neual netwok-based method to pefom high-pecision, obust
More informationPOMDP: Introduction to Partially Observable Markov Decision Processes Hossein Kamalzadeh, Michael Hahsler
POMDP: Intoduction to Patially Obsevable Makov Decision Pocesses Hossein Kamalzadeh, Michael Hahsle 2019-01-02 The R package pomdp povides an inteface to pomdp-solve, a solve (witten in C) fo Patially
More informationAn Assessment of the Efficiency of Close-Range Photogrammetry for Developing a Photo-Based Scanning Systeminthe Shams Tabrizi Minaret in Khoy City
Austalian Jounal of Basic and Applied Sciences, 5(1): 80-85, 011 ISSN 1991-8178 An Assessment of the Efficiency of Close-Range Photogammety fo Developing a Photo-Based Scanning Systeminthe Shams Tabizi
More informationAutomatic Classification of Objects in 3D Laser Range Scans
Automatic Classification of Objects in 3D Laser Range Scans Andreas Nüchter, Hartmut Surmann, Joachim Hertzberg Fraunhofer Institute for Autonomous Intelligent Systems (AIS) Schloss Birlinghoven D-53754
More informationANALYTIC PERFORMANCE MODELS FOR SINGLE CLASS AND MULTIPLE CLASS MULTITHREADED SOFTWARE SERVERS
ANALYTIC PERFORMANCE MODELS FOR SINGLE CLASS AND MULTIPLE CLASS MULTITHREADED SOFTWARE SERVERS Daniel A Menascé Mohamed N Bennani Dept of Compute Science Oacle, Inc Geoge Mason Univesity 1211 SW Fifth
More informationCommunication vs Distributed Computation: an alternative trade-off curve
Communication vs Distibuted Computation: an altenative tade-off cuve Yahya H. Ezzeldin, Mohammed amoose, Chistina Fagouli Univesity of Califonia, Los Angeles, CA 90095, USA, Email: {yahya.ezzeldin, mkamoose,
More informationExtract Object Boundaries in Noisy Images using Level Set. Final Report
Extact Object Boundaies in Noisy Images using Level Set by: Quming Zhou Final Repot Submitted to Pofesso Bian Evans EE381K Multidimensional Digital Signal Pocessing May 10, 003 Abstact Finding object contous
More informationPerformance Optimization in Structured Wireless Sensor Networks
5 The Intenational Aab Jounal of Infomation Technology, Vol. 6, o. 5, ovembe 9 Pefomance Optimization in Stuctued Wieless Senso etwoks Amine Moussa and Hoda Maalouf Compute Science Depatment, ote Dame
More informationUniversity of Alberta, range data with the aid of an o-the-shelf video-camera.
Single Camea Steeo fo Mobile Robot Wold Exploation Dmity O. Goodnichy and William W. Amstong Depatment of Computing Science, Univesity of Albeta, Edmonton, Albeta, Canada T6G 2H1 fdmiti,amsg@cs.ualbeta.ca
More informationXFVHDL: A Tool for the Synthesis of Fuzzy Logic Controllers
XFVHDL: A Tool fo the Synthesis of Fuzzy Logic Contolles E. Lago, C. J. Jiménez, D. R. López, S. Sánchez-Solano and A. Baiga Instituto de Micoelectónica de Sevilla. Cento Nacional de Micoelectónica, Edificio
More informationAssessment of Track Sequence Optimization based on Recorded Field Operations
Assessment of Tack Sequence Optimization based on Recoded Field Opeations Matin A. F. Jensen 1,2,*, Claus G. Søensen 1, Dionysis Bochtis 1 1 Aahus Univesity, Faculty of Science and Technology, Depatment
More informationThe KCLBOT: Exploiting RGB-D Sensor Inputs for Navigation Environment Building and Mobile Robot Localization
The KCBOT: Exploiting GB-D Senso Inputs fo Navigation Envionment Building and Mobile obot ocalization egula Pape Evangelos Geogiou 1,*, Jian Dai 1 and Michael uck 1 1 King s College ondon *Coesponding
More informationMulti-azimuth Prestack Time Migration for General Anisotropic, Weakly Heterogeneous Media - Field Data Examples
Multi-azimuth Pestack Time Migation fo Geneal Anisotopic, Weakly Heteogeneous Media - Field Data Examples S. Beaumont* (EOST/PGS) & W. Söllne (PGS) SUMMARY Multi-azimuth data acquisition has shown benefits
More informationPrioritized Traffic Recovery over GMPLS Networks
Pioitized Taffic Recovey ove GMPLS Netwoks 2005 IEEE. Pesonal use of this mateial is pemitted. Pemission fom IEEE mu be obtained fo all othe uses in any cuent o futue media including epinting/epublishing
More informationAccurate Diffraction Efficiency Control for Multiplexed Volume Holographic Gratings. Xuliang Han, Gicherl Kim, and Ray T. Chen
Accuate Diffaction Efficiency Contol fo Multiplexed Volume Hologaphic Gatings Xuliang Han, Gichel Kim, and Ray T. Chen Micoelectonic Reseach Cente Depatment of Electical and Compute Engineeing Univesity
More informationCOLOR EDGE DETECTION IN RGB USING JOINTLY EUCLIDEAN DISTANCE AND VECTOR ANGLE
COLOR EDGE DETECTION IN RGB USING JOINTLY EUCLIDEAN DISTANCE AND VECTOR ANGLE Slawo Wesolkowski Systems Design Engineeing Univesity of Wateloo Wateloo (Ont.), Canada, NL 3G s.wesolkowski@ieee.og Ed Jenigan
More informationEffective Data Co-Reduction for Multimedia Similarity Search
Effective Data Co-Reduction fo Multimedia Similaity Seach Zi Huang Heng Tao Shen Jiajun Liu Xiaofang Zhou School of Infomation Technology and Electical Engineeing The Univesity of Queensland, QLD 472,
More informationEnvironment Mapping. Overview
Envionment Mapping 1 Oveview Intoduction Envionment map constuction sphee mapping Envionment mapping applications distant geomety eflections 2 1 Oveview Intoduction Envionment map constuction sphee mapping
More informationA VECTOR PERTURBATION APPROACH TO THE GENERALIZED AIRCRAFT SPARE PARTS GROUPING PROBLEM
Accepted fo publication Intenational Jounal of Flexible Automation and Integated Manufactuing. A VECTOR PERTURBATION APPROACH TO THE GENERALIZED AIRCRAFT SPARE PARTS GROUPING PROBLEM Nagiza F. Samatova,
More informationDesired Attitude Angles Design Based on Optimization for Side Window Detection of Kinetic Interceptor *
Poceedings of the 7 th Chinese Contol Confeence July 6-8, 008, Kunming,Yunnan, China Desied Attitude Angles Design Based on Optimization fo Side Window Detection of Kinetic Intecepto * Zhu Bo, Quan Quan,
More information