Disagreement-Based Multi-System Tracking

Size: px
Start display at page:

Download "Disagreement-Based Multi-System Tracking"

Transcription

1 Dsagreement-Based Mult-System Trackng Quannan L 1, Xnggang Wang 2, We Wang 3, Yuan Jang 3, Zh-Hua Zhou 3, Zhuowen Tu 1 1 Lab of Neuro Imagng, Unversty of Calforna, Los Angeles 2 Huazhong Unversty of Scence and Technology 3 Natonal Key Laboratory for Novel Software Technology, Nanjng Unversty Abstract. In ths paper, we tackle the trackng problem from a fuson angle and propose a dsagreement-based approach. Whle most exstng fuson-based trackng algorthms work on dfferent features or parts, our approach can be bult on top of nearly any exstng trackng systems by explotng ther dsagreements. In contrast to assumng mult-vew features or dfferent tranng samples, we utlze exstng well-developed trackng algorthms, whch themselves demonstrate ntrnsc varatons due to ther desgn dfferences. We present encouragng expermental results as well as theoretcal justfcaton of our approach. On a set of benchmark vdeos, large mprovements (20% 40%) over the state-ofthe-art technques have been observed. 1 Introducton Object trackng has been a long standng problem n vson. Once a tracker gets ntalzed, t starts to track the target n a vdeo by performng two steps: (1) makng a predcton about the locaton of the target, and (2) updatng ts object model (locaton, appearance, and shape) based on the predcton. Ths s n sprt very smlar to the bootstrappng and learnng procedure n a learnng algorthm. Wth the recent success n detecton-based trackng approaches, an ncreasng amount of work has treated the trackng problem as a sem-supervsed learnng problem [1 5]. Pckng a target to track at the begnnng provdes supervsed data; the remanng of the frames for the tracker to explore do not contan label nformaton and thus s unsupervsed. Due to the errors ntroduced n both the predcton and model updatng stage, nearly any tracker wll eventually fal wth the errors beng accumulated over the tme. Dsagreement-based sem-supervsed learnng approaches [6], such as co-tranng or tr-tranng [7, 8], provde a mechansm to allow classfers traned on dfferent vews or data samples to explot unlabeled data. The learnng process s a type of ensemble learnng [9 11]. It nvolves multple classfers whch label the unlabeled data to update and mprove each other [12]. From a dfferent angle, the use of multple classfers can be vewed as a fuson problem and t has been shown that fusng complementary features n a trackng system often leads to enhanced performances [13 15]. However, less efforts have been made n learnng to fuse well-developed exstng algorthms through sem-supervsed learnng; we wll see

2 2 Quannan L, et.al. later (n both theory and experments) that a dsagreement-based fuson sgnfcantly mproves the performance over drect combnaton of features/systems [14, 16, 17]. In ths dsagreement-based mult-system trackng approach, we seek a balance between the current tracker and the level of agreements among other trackers. Our ntuton s to fnd the locaton where the current tracker s confdent but dsagrees wth other trackers, whle other trackers reach a hgh degree of agreement. We provde both theoretcal and expermental evdence to our approach and show much mproved results over the state-of-the-art technques on benchmark vdeos. 2 Related Work A number of trackng methods have been proposed to perform fuson[13,14,18, 19, 16, 15, 17]. Dfferent from [13, 17] where multple parts were tracked and correlated, we deal wth a sngle target. In [14, 16] multple trackers were fused but these trackers represent dfferent features and they were drectly combned. In [18] the trackng approach was combned va the weghted combnaton of the PDFs. Dfferent from [18], our method does not perform drect multplcaton but seeks a balance between the PDF of one tracker and the degree of agreement by the other trackers; also, n our method, each tracker performs predcton separately mantanng certan ndependence and patches at the agreed postons can be recommended to update the other trackers. In [20], the trackng combnaton method s traned for specfc scenaros. Dfferent from [20], our method s based on the dsagreement-based sem-supervsed learnng and do not requre an off-lne tranng process; also, t can be appled to general vdeos, and performs very well on a farly large benchmark dataset. In [21], mutual nformaton was used for the fuson. Here, the proposed fuson approach s based on the dsagreements among the trackers. The most related work to our approach s [2], where the co-tranng dea was used to retran classfcaton-based trackers. However, [2] followed the standard co-tranng mplementaton usng one specfc type of classfer, SVM. In [19] several trackng algorthms were combned n a Bayesan framework whereas we here emphasze dsagreement-based fuson through sem-supervsed learnng. In dsagreement-based sem-supervsed learnng, much of the work has been focused on usng mult-vew features [7] or dfferent data samples [8]. The sprt of all such knd of approaches [7,22,8, 23] s to tran multple classfers wth dsagreements, and then label the unlabeled nstances for each other to update/mprove the model. [24] provded PAC bounds wth mult-vew features, whle [12] provded a suffcent condton for mult-vew as well as sngle-vew features. Recently, a suffcent and necessary condton was proved for dsagreementbased sem-supervsed learnng, by establshng a connecton between dsagreementbased and graph-based approaches [25]. In ths paper, we emphasze takng advantages of havng varous well-developed trackng algorthms. In the democratc co-learnng framework [26], dfferent al-

3 Dsagreement-Based Mult-System Trackng 3 gorthms are also used; however, ther approach s for classfcaton and a drect votng of all the methods s used. A man dfference between trackng and classfcaton s that there s no labelng nformaton provded once the trackng process starts (t s a dynamc system), whereas most dsagreement-based semsupervsed learnng algorthms can stll use labeled data n retranng. Notce that the exstence of large dsagreements among the classfers s a premse for the learnng or trackng process to contnue [12], whle the predcton s made by seekng the agreements among the classfers. For example, n [22, 23], classfers are learned so that they not only ft the supervsed data well, but also themselves reach a hgh degree of agreement; the tr-tranng algorthm [8] uses confdent and agreed data from two classfers to help the thrd classfer. Our agreement s used n the predcton stage lke the bootstrappng stage n [26], and we further emphasze the consstency wth the nformaton provded by the current tracker. Our work s also related to the actve learnng lterature [27] but we do not have humans n the loop. 3 Dsagreement-Based Trackng The problem of makng predctons n a trackng system has ts own unque characterstc, and drectly applyng the standard co-tranng formulaton [2] may not necessarly yeld a good soluton. Instead, we take advantages of havng well-developed exstng algorthms, experts, and combne them by explotng the dsagreements among the experts. The dfferences n the ntrnsc desgn of the exstng systems wll naturally lead to a certan amount of bases/varatons, a property the dsagreement-based approaches requres [12]. 3.1 Predcton of Sngle Tracker In ths secton, we frst clarfy our notatons for a sngle tracker. A tracker can be vewed as a learner denoted by h t = (A, f t, X t ) snce t always updates tself. Here, A s a specfc trackng method e.g. mean-shft tracker [28], or partcle flterng [29]; f t s the underlyng appearance model about the target at tme t whch can be represented by a dscrmnatve model [5], generatve model [4], or template matchng [28]; X t s the poston of the target at tme stamp t. Gven a new mage I t+1 at tme stamp t+1, tracker h t makes a predcton, X t+1, about the poston of the target and updates ts underlyng appearance model to f t+1. We can vew trackng a target of a tracker A as computng q t+1 A (x) p A(y x = +1 I t+1 (x), f t ) p(x t+1 = x, X t ) and x q t+1 A (x) = 1 (1) Here y x = +1 ndcates the occurrence of target at locaton x and I t+1 (x) s an mage patch centered at x. Moton coherence s assumed that the predcton on the tme stamp t + 1 s smooth w.r.t. to the predcton on the tme stamp t,

4 4 Quannan L, et.al. for example, p(x t+1 = x, X t ) can be a constant wthn a neghborhood of X t and zero outsde. Ths corresponds to the local search strategy adopted by most of the trackers. Now that q t+1 A (x) [0, 1] ndcates how lkely x t+1 s the correct poston for the target. For an exstng trackng algorthm, t may not strctly follow the formulaton as n Eq. (1), but we stll can use t so long as t outputs a probablty map for the predcton. 3.2 Dsagreement of Trackers Suppose we have a set of exstng trackers (experts) for makng a predcton n a trackng system S = {h, = 1..n} wth n 3 beng the number of trackers and each h s a tracker traned by trackng algorthm A. Gven an nput I t+1 at a tme stamp t + 1, each tracker h computes a q t+1 (x) to make a predcton of random varable X. Let p t+1 (x) denote the ground probablty map whch ndcates how lkely x s the correct poston, our general objectve s to combne the probablty maps by dfferent trackers to obtan hgh probablty modes n the ground-truth p t+1 (x). A drect way to fuse the multple trackers s by lnearly combnng the probablty maps together [15]. Here, we call t drect tracker fuson (DTF), whch serves as a baselne algorthm: q t+1 (x) = 1 n n =1 q t+1 (x) (2) wth the hope that q t+1 (x) p t+1 (x) as each tracker beng unbased and ndependent. Algorthms lke [15] perform n ths way wth an adapton n the weghtng parameters. The target locaton s retreved by ˇx t+1 = arg max x q t+1 (x). In DTF, at each tme, all trackers use the same predcton, ˇx t+1, and each tracker updates ts appearance model to f t+1 based on ˇx t+1 separately and contnues the trackng process. Fusng trackers leads to mprovement over the orgnal ones (see Secton 3.2 and Secton 4 for theoretcal and emprcal justfcaton respectvely). However, predctng the poston for the target w.r.t. Eq. (2) has a bg drawback,.e., the average performance q t+1 (x) of the n trackers may be degenerated by one bad tracker n the group. Here we gve an example to llustrate ths: suppose there are four trackers f 1, f 2, f 3, f 4 and two canddate postons x and x at t+1, where x s the correct poston for the target. The outputs of the four trackers on the two canddate postons are q 1 (x ) = 0.9, q 2 (x ) = 0.4, q 3 (x ) = 0.4, q 4 (x ) = 0.4 and q 1 (x ) = 0.1, q 2 (x ) = 0.6, q 3 (x ) = 0.6, q 4 (x ) = 0.6. The tracker f 1 s very confdent but dsagrees wth other trackers and makes a wrong predcton. To some extent, ths knd of tracker whch s confdent but dsagrees wth other trackers can be thought of as a outler tracker. If we fuse the four trackers wth drect tracker fuson (DTF), the poston x wll be predcted as the poston for the target accordng to x = arg max x q t+1 (x). Unfortunately,

5 Dsagreement-Based Mult-System Trackng 5 we get a wrong poston x due to the outler tracker f 1 at t + 1 although three trackers make correct predcton wth confdence larger than 0.5. Let ζ t+1 denote the probablty mass on such an event that the average performance q t+1 (x) s degenerated by some outler tracker f,.e., the other n 1 trackers agree wth each other and predct the correct poston wth hgh confdence whle the DTF n Eq. (2) predct the poston wrongly due to the outler tracker f at t + 1. Next, we gve the formulaton for combnng the multple trackers based on ther dsagreement to avod ths knd of event for the purpose of robustness. Gven n trackers, we stll let each tracker perform predcton separately. If the current tracker s confdent but dsagrees wth other trackers whle other trackers reach a hgh degree of agreement, the current tracker s prone to be drfted to the agreed poston of other trackers to reach more robust predctons. Our ntuton s that we seek a balance between the generated dstrbuton q t+1 (x) of the current tracker and the degree of agreement by the other trackers as (x) = (1 α)q t+1 (x) + α n 1 [ δ( j, qj t+1 (x) TH) n j=1,j q t+1 j (x)] and the specfc locaton by the -th tracker s x t+1 = arg max x (x). TH s a threshold correspondng to a confdence zone. α balances the mportance of each tracker s own predcton and the nfluence from other trackers. The dervaton of TH and α wll be gven n Secton 3.2. Note that the second term s non-zero only when all the other trackers have a hgh-degree agreement; ths s dfferent from the tradtonal fuson-based trackng [15] where weghted sum s performed; n addton, we emphasze that Eq. (3) focuses manly on the places wth hgh probablty and t s not necessary to ft p t+1 (x) at all xs as n the general statstcal learnng; our dsagreement formulaton n Eq. (3) can take advantage of ths property. Eq. (3) can be understood as the followng: f the current tracker dsagree wth the other trackers whle the other trackers are confdent and agree wth each other, the predcton of the current tracker wll be nfluenced towards the agreed locaton (dependng upon the overall probablty map); otherwse, tracker h gves out a predcton as f there were no other trackers. In such a way, the trackers can keep relatve ndependence and also enable confdent nteractons between each other. Ths makes our approach robust to outler trackers. In addton, usng the agreement of other trackers gves the overall system an ablty to be self-aware of when the system starts to drft. Ths happens when all trackers have hgh entropy of q t+1 (x) wth large dsagreement. The overall output s then gven by x t+1 = argmax x (x) and (x) = 1 n n =1 (3) (x) (4)

6 6 Quannan L, et.al. Note that x t+1 s the output of the overall system but t does not partcpate n the retranng of the ndvdual trackers. The pseudo code of dsagreementbased trackng s shown n Fg. 1. Trackng based on the dsagreements among the trackers shows advantage over usng a drect combnaton and we justfy ths pont both theoretcally and emprcally n the followng sectons. Gven n trackers {h, = 1..n}, each tracker h = (A, f t, X t ) adopts a specfc trackng method A. At the tme stamp t = 0, a target s manually dentfed located at X 0. All trackers start wth the same X 0 and obtan ther appearance model f 0. Gven a new mage I t+1 at tme stamp t + 1, Each tracker h searches a local neghborhood around X t and generate a probablty map q t+1 usng Eq. (1). Fnd modes of x t+1 for (x) as n Eq. (3). x t+1 keeps a balance between the estmaton of the current tracker and the level of agreements among other trackers. Assgn X t+1 = x t+1, sample patches around X t+1 and update the appearance model of each tracker to f t+1 usng the embedded model updatng/learnng rule n A Based on x t+1 = arg max x Qt+1 (x), report the x t+1 as the trackng result for dsagreement-based trackng (DBT). Fg.1. Pseudo code of dsagreement-based trackng. Theoretcal Justfcaton We frst show that a lnear combnaton of multple trackers as n Eq. (2), drect tracker fuson (DTF), gans mprovement over the ndvdual systems. Let p t+1 (x) denote the ground truth whch ndcates how lkely x s the correct poston, and let q t+1 (x) [0, 1] be the output of algorthm A. Lemma 1. If we take an average of the predctons from all the experts: q t+1 (x) = 1 n n =1 qt+1 (x) as n Eq. (2), then the average s bounded n a PAC sense. We suppose that the n trackers are ndependent and unbased: then q t+1 (x) p t+1 (x) as n +. Proof. For any small ǫ > 0, wth Hoeffdng nequalty, we get that P( q t+1 (x) p t+1 (x) ǫ) 2 exp( 2nǫ 2 ). Lemma 1 shows that q t+1 (x) can converge to the ground truth p t+1 (x) exponentally. Let error t+1 denote the error rate of the tracker f at t + 1,.e., the probablty that f predcts a wrong poston for the target at t + 1, error t+1 mn = mn {error t+1 } and errormax t+1 = max {error t+1 }, we gve the followng theorem to show that fusng the multple trackers accordng to Eq. (3) and Eq. (4) wll mprove the performance at least ζ t+1 n(errormax) t+1 n 1, contrastng to the drect tracker fuson (ζ t+1 was defned n the prevous secton).

7 Dsagreement-Based Mult-System Trackng 7 Theorem 1. If we fuse the multple trackers accordng to Eq. (3) and Eq. (4) where α 2 3 and TH 1 2, contrastng to the drect tracker fuson n Eq. (2), the performance at t+1 can be mproved at least ζ t+1 n(errormax) t+1 n 1, where ζ t+1 s the probablty of the event that the average performance q t+1 (x) s degenerated by some outler tracker. Proof. Let X t+1 denote the set of canddate postons at t + 1. If there s some x X t+1, at whch q t+1 (x ) TH for all {1,...,n}, t s easy to fnd that such x s unque, snce TH 0.5 (Here we neglect the probablty mass on the event that at t + 1 there are two postons x and x at whch q t+1 (x) = q t+1 (x ) = 1 2 ). Consderng Eq. (3) we get Qt+1 (x ) = 1 n n k=1 qt+1 k (x ), and x wll be selected as the trackng result, no matter whether argmax (x) or argmax q t+1 s used. If for any x X t+1 there are less than n 1 trackers wth q t+1 (x) TH, then the second term of Eq. (3) s zero. So for any x X t+1, (x) = 1 α n n k=1 qt+1 k (x). Predctng the tracng result accordng to arg max (x) s equal to predctng accordng to argmax q t+1. Next we analyze the stuaton when there s some x X t+1, at whch q t+1 ( x) < TH and q t+1 j ( x) TH for all j. Obvously, such x s also unque, snce TH 0.5 and n 3. Case 1: x s the correct poston for the target at t + 1. We wll show that even f q t+1 ( x) s very close to 0,.e., tracker f s an outler at t + 1, t wll not degenerate the fuson of the multple trackers due to Eq. (3). We obtan Thus, ( x) = (1 α)q t+1 ( x) + α n 1 j,j n k=1 j q t+1 j ( x) (1 α)q t+1 ( x) + α TH (5) ( x) = (1 α)qt+1 j ( x) (1 α) TH k ( x) (1 α)q t+1 ( x) + α TH + (1 α)(n 1)TH (6) For an ncorrect poston x x, snce x X q t+1 t+1 k (x) = 1, t s easy to see that (x ) = (1 α)q t+1 (x ) (1 α)(1 q t+1 ( x)) Therefore, j,j (x ) = (1 α)qj t+1 (x ) (1 α)(1 q t+1 ( x)) (1 α)(1 TH) (7) n k=1 k (x ) (1 α)(1 q t+1 ( x)) + (1 α)(n 1)(1 TH) (8)

8 8 Quannan L, et.al. We see that n general n k=1 Qt+1 k ( x) n k=1 Qt+1 k (x ) for α 2 3 and TH 1 2. Ths makes the correct poston more robust,.e., the predcton of the dsagreementbased trackng wll never be nfluenced even f f s a outler tracker. So the mprovement s at least ζ t+1. Case 2: x s not the correct poston for the target at t + 1. Snce q t+1 j ( x) TH for all j and TH 1 2, n 1 trackers predct the wrong poston x as the trackng result at t + 1. Now we bound the probablty of such event. Snce error t+1 j errormax t+1 and the multple trackers are assumed to be ndependent, the probablty mass on the event that n 1 trackers predct the poston mstakenly s at most n(errormax t+1 )n 1. The worst stuaton s that the fuson accordng to Eq. (3) performs worse than the drect tracker fuson n case 2 completely. We get Theorem 1 proved. From Theorem 1 we know that the fuson wll get beneft from Eq. (3) under the stuaton that one bad tracker degenerates the drect tracker fuson. When n (the number of the trackers) s large, t would be dffcult for the remanng n 1 trackers to acheve some agreement (See the experment Non-Relax n Secton 4). In practce, we can relax ths constrant, e.g., when two or more trackers acheve agreement, the agreement term would take effect. Note that the performance of Eq. (2) and Eq. (3) depends on the correlaton between the trackers. The correlaton depends on two factors: (1) the ntrnsc desgn of the trackers; (2) the tranng samples used to tran the trackers. Two trackers wth the same desgn traned on the same set of samples are hghly correlated, and two dfferent types of trackers traned on the same set of samples are more correlated than those traned on dfferent set of samples. If the n trackers are the same, then usng Eq. (3) shows no advantage over Eq. (2). In summary, Lemma 1 suggests that fusng the multple experts drectly mght gan exponental mprovement, contrastng to the sngle tracker; Theorem 1 shows that our dsagreement-based fuson method can provde more robustness to the trackng system, whch motvates the use of Eq. (3) by keepng a balance between the current expert f and the agreement from the other experts. 4 Experments In the experments, we make a comprehensve comparson between the performance of dsagreement-based trackng, drect tracker fuson, and the ndvdual trackers. Four trackers are used and the experment s conducted on 11 commonly tested vdeos (lsted n Table 1). The trackers used are MlTracker [5], the semsupervsed on-lne boostng tracker (semboost)[3], Incremental Vsual Tracker (IVT)[4], and Incremental Vsual Tracker usng edge nformaton (IVTE). Snce the ndvdual trackers perform predcton separately, the computatonal complexty of the proposed method only adds slght overhead over the ndvdual ones wth the mult-core processor and parallel computng. Compared wth the large performance gan, ths computatonal overhead s tolerable. From the experments, we observe that, statstcally, each ndvdual tracker gets sgnfcantly mproved by usng Eq. (3): the average center locaton error

9 Dsagreement-Based Mult-System Trackng 9 has been reduced by more than 12 pxels and the success rate has ncreased by 20% 40%. The result of DBT (Dsagreement-Based Trackng) also outperforms the system by drectly combnng the orgnal trackers,.e., DTF, wth 3.3 pxels reducton n center locaton error and 4.4% mprovement n success rate. DBT also sgnfcantly outperforms PROST [30]. 4.1 Implementaton detals Whle a wealthy body of trackng papers/systems have been reported, we found a few systems (wth avalable source code) havng decent performance on general vdeos. Here, we provde bref descrptons for these trackers we used wth necessary changes made to them. MlTracker adopts an onlne multple nstance learnng algorthm to tran a dscrmnatve classfer. In order to handle the ambguty of sampled patches, a bag of potentally postve mage patches are extracted. MlTracker mantans a pool of Haar features and the onlne boostng mechansm s adopted. SemBoostTracker also adopts an onlne boostng mechansm and t formulates the update process n a sem-supervsed fashon combned wth a gven pror. Ths helps to allevate the drftng problem. The IVT ncrementally learns a low dmensonal egenspace representaton to model the appearance changes of the object. In IVT model, the target s represented as a vector of gray-scale value, and the moton s modeled by an affne mage warpng. To propagate sample dstrbutons over tme, a partcle flter framework s adopted. Snce both MILTracker and SemBoostTracker do not support affne transformaton, we dsabled the scalng and rotatng ablty of IVT. IVTE s smlar to IVT, except that t uses level set as the feature. The forms of the 4 trackng systems outputs are rather dfferent. MILTracker and SemBoostTracker produce scores on local search regons; IVT and IVTE propagate probabltes va partcles. In the experment, we map the scores of MILTracker and SemBoostTracker to the range [0, 1] to produce probablty maps q MIL and q SBT (The probablty maps are normalzed to make sure that x qt+1 A (x) = 1). For IVT and IVTE, we keep the poston entres (a, b ) of the partcle and use a parzen wndow approach to estmate the probablty for predcton. For a pont x = (a, b) on the mage, ts probablty s calculated as q IV T (x) = M =1 w max{0, 1 (a a ) 2 + (b b ) 2 /L}. In our experment, L s set to 15. A map q IV TE s produced smlarly as q IV T for IVTE. Based on q MIL, q SBT, q IV T, and q IV TE we respectvely compute the correspondng Q MIL, Q SBT, Q IV T, and Q IV TE usng Eq. (3) (Snce 4 trackers are used, we relaxed Eq. (3) that when 2 trackers acheve confdent agreement, the agreement term wll take effect) and thus, each tracker makes ts own predcton separately. As we have mentoned before, x t+1 found by mean shft algorthm [28] can represent multple ponts (modes). For each tracker, e.g., MILTracker, the one mode wth the maxmum value s reported as ts predcton, sgnfcant modes found are used to retran the tracker and as the seeds for further search at the next tme stamp. For the results reported n our experment, α s set to be 0.67 as suggested by the theoretcal secton. The threshold TH n Eq. (3) s

10 10 Quannan L, et.al. set as 0.8/(Λ/3) (Λ s the sze of the search wndow). For each tracker, 2 modes are kept. 4.2 Quanttatve Results The average center locaton error The average center locaton error s a commonly used metrc to measure the performance of trackng and s defned as the average error between the predcted locatons to the ground truth. In Table 1, we summarze the results of the average center locaton error on all the 11 vdeos. It s clear that our dsagreement based tracker outperforms the ndvdual trackers, Drect tracker fuson, and the co-tranng scheme. For nonrelax (usng Eq. (3) drectly wthout relaxaton), t s less possble for the trackers to acheve some agreements, and the chance of nteracton between trackers s reduced. Stll the result s better than the ndvdual trackers. Table 1. Comparson of Average Center Locaton Error. Non-Relax ndcates to use Eq. (3) drectly wthout relaxaton;co-tranng stands for the results usng co-tranng method. vdeos MlT IVT IVTE SBT DTF Co-Tranng Non-Relax DBT Grl CokeCan Tger Sylv StatOcc Davd Clffbar Surfer faceocc Indoor faceocc In all The success rate If the locaton error on one frame s less than a prespecfed threshold, the predcton s regarded as a successful predcton. The success rate s defned as the rato of successful predctons over all the predctons. To compute the success rate, we set T as certan rato of the average wdth of the target,.e., T = β (w + h)/4, where, w and h are the wdth and heght of the target respectvely. Conceptually, ths s smlar to the overlap score evaluaton used n [30] and thus, success rate s conceptually smlar to the tracked percentle n [30]. We observe that, the trackers can not track the target precsely at all tmes, f there s no overlap but the predcton of the tracker s not far from the target or wthn the search area of the tracker, t s often possble for the trackng process to recover. Our evaluaton measurement can reflect such a phenomenon. We compare the success rate n Table 2 and from Table 2, the DBT acheves the hghest success rate.

11 Dsagreement-Based Mult-System Trackng 11 Table 2. Comparson of success rate when β = 0.5. MlT IVT IVTE SBT DTF Non-Relax DBT Comparson of probablty maps In Fg. 2, we show how the probablty maps are generated n Eq. (2) and Eq. (3) on a testng vdeo, Tger1. DTF drfts from the 30th frame; on ths frame, the predctons of the trackers are rather dverse. In DTF, however, the ndvdual tracker s predcton s not fully respected and t has to comply wth the voted predcton; ths s the prmary reason for drftng and gettng trapped; usng Eq. (3) however leads to a more robust predcton. As we can see from the second fgure, on the 30th frame, MILTracker and Semboost tracker acheves certan agreement, but IVT and IVTE stll comples wth ts own predcton snce the agreement s weak. On the 35th frame, when MILTracker, IVT and Semboost tracker acheve confdent agreement, IVTE s pulled back to the agreed poston and the four trackers merge agan. The beneft of our dsagreement-based trackng s obvous: the trackers then keep ther relatvely dfferent traces and the rsk of gettng trapped s reduced. (a) (b) (c) Fg. 2. Illustraton of the probablty maps where four trackers (experts) are adopted (the fgures have been scaled for vsualzaton). (a) shows the results by DTF. (b) and (c) dsplay the probablty maps generated by dsagreement-based trackng. The probablty maps nsde the dashed yellow rectangle are shown below the screen shots. Underneath each fgure, from left to rght, the probablty maps are IVT, IVTE, MIL- Tracker, Semboost tracker respectvely (see the dscussons about these trackers n the experments). For DTF n (a), the ffth probablty map s the combned map. For dsagreement-based trackng n (b) and (c), the frst rows shows the orgnal probablty maps, and the second row shows the computed by Eq. (3). Comparson wth other methods PROST [30] s another fuson based trackng algorthm that adopts 3 trackers (a template model, an optcal-flow based mean-shft tracker and an onlne random forest tracker). Table 3 and Table 4 compare the average locaton errors and the tracked percentage (computed usng the overlap score n [30]) wth PROST. From the two tables, we can observe that, our dsagreement-based trackng outperforms PROST and acheves better tracked percentages on most of the vdeos. The best expermental performance of Democratc Integraton n [15] was acheved by usng unform qualtes, whch assgned equal weghts to all the

12 12 Quannan L, et.al. Table 3. Comparson of average center locaton error wth [30] Method Grl tger1 sylv Davd faceocc faceocc2 [30] Ours Table 4. Comparson of tracked percentage wth [30] Method Grl tger1 sylv Davd faceocc faceocc2 [30] Ours clues, and corresponded drectly wth DTF. In addton, we mplemented the qualty measure of normalzed salency and the performance was not as good as DBT: ther average center locaton error s 13.8 wth success rate We dd not get the mplementaton of [2]. Nevertheless, we dd experment on some vdeos used n [2] and the results of DBT are better than [2] qualtatvely (skpped here due to page lmt). Moreover, we ndeed mplemented co-tranng and reported the result n Table 1 (average center locaton error 32.7), whch s much worse than DBT. Table 5. Performance by varyng α and T H (T H = R/(Λ/3)) R/α 0.8/ / / / /0.67 Average Error Success Rate Robustness by varyng the parameters In table 5, we summarze the performance of dsagreement-based trackng by varyng the parameters α and TH, whch are the two key parameters n Eq. (3). We can see from ths table, by varyng TH and α, the results (especally the success rate) do not change too much. Ths demonstrates the robustness of dsagreement-based trackng. The average center locaton error has relatvely larger change because on portons of the vdeos Tger1 and Indoor, dsagreement-based trackng gets dstracted to postons dstant from the targets. In such cases, the success rate does not vary too much, but the center locaton error s ncreased. Screenshots of the results In Fg. 3, we compare the trackng results on the vdeo Grl. Ths vdeo undergoes several challenges: fast appearance change and occluson. Although both MILTracker and IVTE can track the face of the grl successfully, the trackng process s not very stable. IVT drfts from the face at the 20th frame. From the 391th frame, drect tracker fuson also drfts and get trapped at the background of the mages. As can be seen from both the screen shots and the error plot on the rght of Fg. 3, we fnd that dsagreementbased trackng tracks most robustly and accurately. In Fg. 3, we can also fnd

13 Dsagreement-Based Mult-System Trackng 13 Fg.3. Comparson of trackng results on the vdeo Grl. The frst row on the left shows the results of dsagreement-based trackng; the second row on the left shows the results of 4 ndvdual trackers and drect tracker fuson; the rght plot shows the comparson of center locaton error (the number of the x axs denotes the number of predctons). The colorng scheme s: dotted green: IVT, dotted black: MILTracker, dotted blue: IVTE, dotted yellow: Semboost tracker, sold Red: dsagreement-based trackng, and sold magenta: Drect tracker fuson. a very nce property of democratc trackng that, the traces of the four trackers are smlar but not exactly the same, thus, they can explore dfferent spaces, recommend confdent samples to other trackers, and thus avod to be trapped at an ncorrect poston. 5 Concluson In ths paper, we have ntroduced a dsagreement-based trackng method whch fuses multple exstng trackng systems n the followng way that seeks a balance between the coherence of the current tracker and the degree of agreements among other trackers. In such a way, t enables the nteracton between trackers and keeps the appealng characterstcs of the trackers at the same tme. As llustrated n the experments, the balance comples wth the characterstc of trackng. Dsagreement-based trackng can be bult on top of varous exstng well-developed trackng systems utlzng ther ntrnsc bases. Adoptng several state-of-the-art trackng algorthms, our approach s able to mprove each of them by a large margn on wdely used benchmark vdeos n the lterature Acknowledgement. Zhuowen Tu and Quannan L are supported by Offce of Naval Research Award N and NSF CAREER award IIS ; We Wang, Yuan Jang and Zh-Hua Zhou are supported by Natonal Natural Scence Fund of Chna References 1. Avdan, S.: Ensemble trackng. In: CVPR. (2005) 2. Tang, F., Brennan, S., Zhao, Q., Tao, H.: Co-trackng usng sem-supervsed support vector machnes. In: ICCV. (2007) 3. Grabner, H., Lestner, C., Bschof, H.: Sem-supervsed on-lne boostng for robust trackng. In: ECCV. (2008) 4. Lm, J., Ross, D., Ln, R.S., Yang, M.H.: Incremental learnng for vsual trackng. Int l J. of Comp. Vs. (2008)

14 14 Quannan L, et.al. 5. Babenko, B., Yang, M.H., Belonge, S.: Vsual trackng wth onlne multple nstance learnng. In: CVPR. (2009) 6. Zhou, Z.H., L, M.: Sem-supervsed learnng by dsagreement. Knowledge and Informaton Systems 24 (2010) Blum, A., Mtchell, T.: Combnng labeled and unlabeled data wth co-tranng. In: COLT. (1998) Zhou, Z.H., L, M.: Tr-tranng: Explotng unlabeled data usng three classfers. IEEE Trans. on Know. and Data Eng. (2005) 9. Detterch, T.G.: Ensemble methods n machne learnng. In: INTERNATIONAL WORKSHOP ON MULTIPLE CLASSIFIER SYSTEMS, Sprnger-Verlag (2000) Kuncheva, L.I.: A theoretcal study on sx classfer fuson strateges. IEEE Tran. on PAMI 24 (2002) Bennett, K.P., Demrz, A., Macln, R.: Explotng Unlabeled Data n Ensemble Methods. In: ACM SIGKDD 02 Edmonton, Alberta CA. (2002) 12. Wang, W., Zhou, Z.H.: Analyzng co-tranng style algorthms. In: ECML. (2007) 13. Wu, Y., Huang, T.: Robust vsual trackng by ntegratng multple cues based on co-nference learnng. In l J. of Comp. Vs (2004) 14. Sebel, N., Maybank, S.: Fuson of multple trackng algorthms for robust people trackng. In: ECCV. (2002) 15. Tresch, J., von der Naksvyrg, C.: Democratc ntegraton: Self-organzed ntegraton of adaptve vsual cues. In: Neuroal Computaton. (2001) 16. Spengler, M., Schele, B.: Towards robust mult-cue ntegraton of vsual trackng. In: MVA. (2003) 17. Kwon, J., Lee, K.M.: Vsual trackng decomposton. In: CVPR. (2010) 18. I Lechter, M Lndenbaum, E.R.: A general framework for combnng vsual trackersthe black boxes approach. It l J. of Comp. Vs (2006) 19. Zhong, B., Yao, H., Chen, S., J, R.R., Yuan, X., Lu, S., Gao, W.: Vsual trackng va weakly supervsed learnng from multple mperfect oracles. In: CVPR. (2010) 20. Stenger, B., Woodley, T., Cpolla, R.: Learnng to track wth multple observers. In: CVPR. (2009) 21. Mundy, J.L., Chang, C.F.: Fuson of ntensty, texture, and color n vdeo trackng based on mutual nformaton. In: AIPR. (2010) 22. Collns, M., Snger, Y.: Unsupervsed models for named entty classfcaton. In: EMNLP. (1999) 23. Leskes, B., Torenvlet, L.: The value of agreement a new boostng algorthm. J. of Comp. and Sys. Sc. (2008) 24. Dasgupta, S., Lttman, M.L., McAllester, D.: Pac generalzaton bounds for cotranng. In: NIPS. (1999) 25. Wang, W., Zhou, Z.H.: A new analyss of co-tranng. In: ICML, Hafa, Israel (2010) Zhou, Y., Goldman, S.: Democratc co-learnng. In: Inte l Conf. on Tools wth Art. Intell. (2004) 27. Dasgupta, S.: Coarse sample complexty bounds for actve learnng. In: NIPS. (2005) 28. Comancu, D., Ramesh, V., Meer, P.: Real-tme trackng of non-rgd objects usng mean shft. In: CVPR. (2000) 29. Isard, M., Blake, A.: Condensaton - condtonal densty propagaton for vsual trackng. Int l J. on Comp. Vs. (1998) 30. Santner, J., Lestner, C., Saffar, A., Pock, T., Bschof, H.: Prost: Parallel robust onlne smple trackng. In: CVPR. (2010)

benefit is 2, paid if the policyholder dies within the year, and probability of death within the year is ).

benefit is 2, paid if the policyholder dies within the year, and probability of death within the year is ). REVIEW OF RISK MANAGEMENT CONCEPTS LOSS DISTRIBUTIONS AND INSURANCE Loss and nsurance: When someone s subject to the rsk of ncurrng a fnancal loss, the loss s generally modeled usng a random varable or

More information

What is Candidate Sampling

What is Candidate Sampling What s Canddate Samplng Say we have a multclass or mult label problem where each tranng example ( x, T ) conssts of a context x a small (mult)set of target classes T out of a large unverse L of possble

More information

Module 2 LOSSLESS IMAGE COMPRESSION SYSTEMS. Version 2 ECE IIT, Kharagpur

Module 2 LOSSLESS IMAGE COMPRESSION SYSTEMS. Version 2 ECE IIT, Kharagpur Module LOSSLESS IMAGE COMPRESSION SYSTEMS Lesson 3 Lossless Compresson: Huffman Codng Instructonal Objectves At the end of ths lesson, the students should be able to:. Defne and measure source entropy..

More information

An Alternative Way to Measure Private Equity Performance

An Alternative Way to Measure Private Equity Performance An Alternatve Way to Measure Prvate Equty Performance Peter Todd Parlux Investment Technology LLC Summary Internal Rate of Return (IRR) s probably the most common way to measure the performance of prvate

More information

PSYCHOLOGICAL RESEARCH (PYC 304-C) Lecture 12

PSYCHOLOGICAL RESEARCH (PYC 304-C) Lecture 12 14 The Ch-squared dstrbuton PSYCHOLOGICAL RESEARCH (PYC 304-C) Lecture 1 If a normal varable X, havng mean µ and varance σ, s standardsed, the new varable Z has a mean 0 and varance 1. When ths standardsed

More information

Forecasting the Direction and Strength of Stock Market Movement

Forecasting the Direction and Strength of Stock Market Movement Forecastng the Drecton and Strength of Stock Market Movement Jngwe Chen Mng Chen Nan Ye cjngwe@stanford.edu mchen5@stanford.edu nanye@stanford.edu Abstract - Stock market s one of the most complcated systems

More information

Logistic Regression. Lecture 4: More classifiers and classes. Logistic regression. Adaboost. Optimization. Multiple class classification

Logistic Regression. Lecture 4: More classifiers and classes. Logistic regression. Adaboost. Optimization. Multiple class classification Lecture 4: More classfers and classes C4B Machne Learnng Hlary 20 A. Zsserman Logstc regresson Loss functons revsted Adaboost Loss functons revsted Optmzaton Multple class classfcaton Logstc Regresson

More information

An Interest-Oriented Network Evolution Mechanism for Online Communities

An Interest-Oriented Network Evolution Mechanism for Online Communities An Interest-Orented Network Evoluton Mechansm for Onlne Communtes Cahong Sun and Xaopng Yang School of Informaton, Renmn Unversty of Chna, Bejng 100872, P.R. Chna {chsun,yang}@ruc.edu.cn Abstract. Onlne

More information

The Development of Web Log Mining Based on Improve-K-Means Clustering Analysis

The Development of Web Log Mining Based on Improve-K-Means Clustering Analysis The Development of Web Log Mnng Based on Improve-K-Means Clusterng Analyss TngZhong Wang * College of Informaton Technology, Luoyang Normal Unversty, Luoyang, 471022, Chna wangtngzhong2@sna.cn Abstract.

More information

Recurrence. 1 Definitions and main statements

Recurrence. 1 Definitions and main statements Recurrence 1 Defntons and man statements Let X n, n = 0, 1, 2,... be a MC wth the state space S = (1, 2,...), transton probabltes p j = P {X n+1 = j X n = }, and the transton matrx P = (p j ),j S def.

More information

Face Verification Problem. Face Recognition Problem. Application: Access Control. Biometric Authentication. Face Verification (1:1 matching)

Face Verification Problem. Face Recognition Problem. Application: Access Control. Biometric Authentication. Face Verification (1:1 matching) Face Recognton Problem Face Verfcaton Problem Face Verfcaton (1:1 matchng) Querymage face query Face Recognton (1:N matchng) database Applcaton: Access Control www.vsage.com www.vsoncs.com Bometrc Authentcaton

More information

1 Example 1: Axis-aligned rectangles

1 Example 1: Axis-aligned rectangles COS 511: Theoretcal Machne Learnng Lecturer: Rob Schapre Lecture # 6 Scrbe: Aaron Schld February 21, 2013 Last class, we dscussed an analogue for Occam s Razor for nfnte hypothess spaces that, n conjuncton

More information

Vision Mouse. Saurabh Sarkar a* University of Cincinnati, Cincinnati, USA ABSTRACT 1. INTRODUCTION

Vision Mouse. Saurabh Sarkar a* University of Cincinnati, Cincinnati, USA ABSTRACT 1. INTRODUCTION Vson Mouse Saurabh Sarkar a* a Unversty of Cncnnat, Cncnnat, USA ABSTRACT The report dscusses a vson based approach towards trackng of eyes and fngers. The report descrbes the process of locatng the possble

More information

The Greedy Method. Introduction. 0/1 Knapsack Problem

The Greedy Method. Introduction. 0/1 Knapsack Problem The Greedy Method Introducton We have completed data structures. We now are gong to look at algorthm desgn methods. Often we are lookng at optmzaton problems whose performance s exponental. For an optmzaton

More information

DEFINING %COMPLETE IN MICROSOFT PROJECT

DEFINING %COMPLETE IN MICROSOFT PROJECT CelersSystems DEFINING %COMPLETE IN MICROSOFT PROJECT PREPARED BY James E Aksel, PMP, PMI-SP, MVP For Addtonal Informaton about Earned Value Management Systems and reportng, please contact: CelersSystems,

More information

Luby s Alg. for Maximal Independent Sets using Pairwise Independence

Luby s Alg. for Maximal Independent Sets using Pairwise Independence Lecture Notes for Randomzed Algorthms Luby s Alg. for Maxmal Independent Sets usng Parwse Independence Last Updated by Erc Vgoda on February, 006 8. Maxmal Independent Sets For a graph G = (V, E), an ndependent

More information

Institute of Informatics, Faculty of Business and Management, Brno University of Technology,Czech Republic

Institute of Informatics, Faculty of Business and Management, Brno University of Technology,Czech Republic Lagrange Multplers as Quanttatve Indcators n Economcs Ivan Mezník Insttute of Informatcs, Faculty of Busness and Management, Brno Unversty of TechnologCzech Republc Abstract The quanttatve role of Lagrange

More information

The OC Curve of Attribute Acceptance Plans

The OC Curve of Attribute Acceptance Plans The OC Curve of Attrbute Acceptance Plans The Operatng Characterstc (OC) curve descrbes the probablty of acceptng a lot as a functon of the lot s qualty. Fgure 1 shows a typcal OC Curve. 10 8 6 4 1 3 4

More information

CS 2750 Machine Learning. Lecture 3. Density estimation. CS 2750 Machine Learning. Announcements

CS 2750 Machine Learning. Lecture 3. Density estimation. CS 2750 Machine Learning. Announcements Lecture 3 Densty estmaton Mlos Hauskrecht mlos@cs.ptt.edu 5329 Sennott Square Next lecture: Matlab tutoral Announcements Rules for attendng the class: Regstered for credt Regstered for audt (only f there

More information

Vehicle Detection and Tracking in Video from Moving Airborne Platform

Vehicle Detection and Tracking in Video from Moving Airborne Platform Journal of Computatonal Informaton Systems 10: 12 (2014) 4965 4972 Avalable at http://www.jofcs.com Vehcle Detecton and Trackng n Vdeo from Movng Arborne Platform Lye ZHANG 1,2,, Hua WANG 3, L LI 2 1 School

More information

Realistic Image Synthesis

Realistic Image Synthesis Realstc Image Synthess - Combned Samplng and Path Tracng - Phlpp Slusallek Karol Myszkowsk Vncent Pegoraro Overvew: Today Combned Samplng (Multple Importance Samplng) Renderng and Measurng Equaton Random

More information

Forecasting the Demand of Emergency Supplies: Based on the CBR Theory and BP Neural Network

Forecasting the Demand of Emergency Supplies: Based on the CBR Theory and BP Neural Network 700 Proceedngs of the 8th Internatonal Conference on Innovaton & Management Forecastng the Demand of Emergency Supples: Based on the CBR Theory and BP Neural Network Fu Deqang, Lu Yun, L Changbng School

More information

Support Vector Machines

Support Vector Machines Support Vector Machnes Max Wellng Department of Computer Scence Unversty of Toronto 10 Kng s College Road Toronto, M5S 3G5 Canada wellng@cs.toronto.edu Abstract Ths s a note to explan support vector machnes.

More information

INVESTIGATION OF VEHICULAR USERS FAIRNESS IN CDMA-HDR NETWORKS

INVESTIGATION OF VEHICULAR USERS FAIRNESS IN CDMA-HDR NETWORKS 21 22 September 2007, BULGARIA 119 Proceedngs of the Internatonal Conference on Informaton Technologes (InfoTech-2007) 21 st 22 nd September 2007, Bulgara vol. 2 INVESTIGATION OF VEHICULAR USERS FAIRNESS

More information

J. Parallel Distrib. Comput.

J. Parallel Distrib. Comput. J. Parallel Dstrb. Comput. 71 (2011) 62 76 Contents lsts avalable at ScenceDrect J. Parallel Dstrb. Comput. journal homepage: www.elsever.com/locate/jpdc Optmzng server placement n dstrbuted systems n

More information

Implementation of Deutsch's Algorithm Using Mathcad

Implementation of Deutsch's Algorithm Using Mathcad Implementaton of Deutsch's Algorthm Usng Mathcad Frank Roux The followng s a Mathcad mplementaton of Davd Deutsch's quantum computer prototype as presented on pages - n "Machnes, Logc and Quantum Physcs"

More information

How Sets of Coherent Probabilities May Serve as Models for Degrees of Incoherence

How Sets of Coherent Probabilities May Serve as Models for Degrees of Incoherence 1 st Internatonal Symposum on Imprecse Probabltes and Ther Applcatons, Ghent, Belgum, 29 June 2 July 1999 How Sets of Coherent Probabltes May Serve as Models for Degrees of Incoherence Mar J. Schervsh

More information

A Probabilistic Theory of Coherence

A Probabilistic Theory of Coherence A Probablstc Theory of Coherence BRANDEN FITELSON. The Coherence Measure C Let E be a set of n propostons E,..., E n. We seek a probablstc measure C(E) of the degree of coherence of E. Intutvely, we want

More information

Traffic State Estimation in the Traffic Management Center of Berlin

Traffic State Estimation in the Traffic Management Center of Berlin Traffc State Estmaton n the Traffc Management Center of Berln Authors: Peter Vortsch, PTV AG, Stumpfstrasse, D-763 Karlsruhe, Germany phone ++49/72/965/35, emal peter.vortsch@ptv.de Peter Möhl, PTV AG,

More information

A Fast Incremental Spectral Clustering for Large Data Sets

A Fast Incremental Spectral Clustering for Large Data Sets 2011 12th Internatonal Conference on Parallel and Dstrbuted Computng, Applcatons and Technologes A Fast Incremental Spectral Clusterng for Large Data Sets Tengteng Kong 1,YeTan 1, Hong Shen 1,2 1 School

More information

Multiple-Period Attribution: Residuals and Compounding

Multiple-Period Attribution: Residuals and Compounding Multple-Perod Attrbuton: Resduals and Compoundng Our revewer gave these authors full marks for dealng wth an ssue that performance measurers and vendors often regard as propretary nformaton. In 1994, Dens

More information

How To Know The Components Of Mean Squared Error Of Herarchcal Estmator S

How To Know The Components Of Mean Squared Error Of Herarchcal Estmator S S C H E D A E I N F O R M A T I C A E VOLUME 0 0 On Mean Squared Error of Herarchcal Estmator Stans law Brodowsk Faculty of Physcs, Astronomy, and Appled Computer Scence, Jagellonan Unversty, Reymonta

More information

A Performance Analysis of View Maintenance Techniques for Data Warehouses

A Performance Analysis of View Maintenance Techniques for Data Warehouses A Performance Analyss of Vew Mantenance Technques for Data Warehouses Xng Wang Dell Computer Corporaton Round Roc, Texas Le Gruenwald The nversty of Olahoma School of Computer Scence orman, OK 739 Guangtao

More information

FREQUENCY OF OCCURRENCE OF CERTAIN CHEMICAL CLASSES OF GSR FROM VARIOUS AMMUNITION TYPES

FREQUENCY OF OCCURRENCE OF CERTAIN CHEMICAL CLASSES OF GSR FROM VARIOUS AMMUNITION TYPES FREQUENCY OF OCCURRENCE OF CERTAIN CHEMICAL CLASSES OF GSR FROM VARIOUS AMMUNITION TYPES Zuzanna BRO EK-MUCHA, Grzegorz ZADORA, 2 Insttute of Forensc Research, Cracow, Poland 2 Faculty of Chemstry, Jagellonan

More information

How To Understand The Results Of The German Meris Cloud And Water Vapour Product

How To Understand The Results Of The German Meris Cloud And Water Vapour Product Ttel: Project: Doc. No.: MERIS level 3 cloud and water vapour products MAPP MAPP-ATBD-ClWVL3 Issue: 1 Revson: 0 Date: 9.12.1998 Functon Name Organsaton Sgnature Date Author: Bennartz FUB Preusker FUB Schüller

More information

Eye Center Localization on a Facial Image Based on Multi-Block Local Binary Patterns

Eye Center Localization on a Facial Image Based on Multi-Block Local Binary Patterns Eye Center Localzaton on a Facal Image Based on Mult-Bloc Local Bnary Patterns Anatoly tn, Vladmr Khryashchev, Olga Stepanova Yaroslavl State Unversty Yaroslavl, Russa anatolyntnyar@gmal.com, vhr@yandex.ru,

More information

Gender Classification for Real-Time Audience Analysis System

Gender Classification for Real-Time Audience Analysis System Gender Classfcaton for Real-Tme Audence Analyss System Vladmr Khryashchev, Lev Shmaglt, Andrey Shemyakov, Anton Lebedev Yaroslavl State Unversty Yaroslavl, Russa vhr@yandex.ru, shmaglt_lev@yahoo.com, andrey.shemakov@gmal.com,

More information

Detecting Global Motion Patterns in Complex Videos

Detecting Global Motion Patterns in Complex Videos Detectng Global Moton Patterns n Complex Vdeos Mn Hu, Saad Al, Mubarak Shah Computer Vson Lab, Unversty of Central Florda {mhu,sal,shah}@eecs.ucf.edu Abstract Learnng domnant moton patterns or actvtes

More information

Can Auto Liability Insurance Purchases Signal Risk Attitude?

Can Auto Liability Insurance Purchases Signal Risk Attitude? Internatonal Journal of Busness and Economcs, 2011, Vol. 10, No. 2, 159-164 Can Auto Lablty Insurance Purchases Sgnal Rsk Atttude? Chu-Shu L Department of Internatonal Busness, Asa Unversty, Tawan Sheng-Chang

More information

Software project management with GAs

Software project management with GAs Informaton Scences 177 (27) 238 241 www.elsever.com/locate/ns Software project management wth GAs Enrque Alba *, J. Francsco Chcano Unversty of Málaga, Grupo GISUM, Departamento de Lenguajes y Cencas de

More information

Sketching Sampled Data Streams

Sketching Sampled Data Streams Sketchng Sampled Data Streams Florn Rusu, Aln Dobra CISE Department Unversty of Florda Ganesvlle, FL, USA frusu@cse.ufl.edu adobra@cse.ufl.edu Abstract Samplng s used as a unversal method to reduce the

More information

Study on Model of Risks Assessment of Standard Operation in Rural Power Network

Study on Model of Risks Assessment of Standard Operation in Rural Power Network Study on Model of Rsks Assessment of Standard Operaton n Rural Power Network Qngj L 1, Tao Yang 2 1 Qngj L, College of Informaton and Electrcal Engneerng, Shenyang Agrculture Unversty, Shenyang 110866,

More information

Combinatorial Fusion Criteria for Real-Time Tracking

Combinatorial Fusion Criteria for Real-Time Tracking Combnatoral Fuson Crtera for Real-Tme Trackng D. Frank Hsu, Daman M. Lyons and Jzhou Computer Vson & Robotcs Laboratory Department of Computer & Informaton Scence Fordham Unversty, Bronx, NY 10458, US

More information

8 Algorithm for Binary Searching in Trees

8 Algorithm for Binary Searching in Trees 8 Algorthm for Bnary Searchng n Trees In ths secton we present our algorthm for bnary searchng n trees. A crucal observaton employed by the algorthm s that ths problem can be effcently solved when the

More information

An Enhanced Super-Resolution System with Improved Image Registration, Automatic Image Selection, and Image Enhancement

An Enhanced Super-Resolution System with Improved Image Registration, Automatic Image Selection, and Image Enhancement An Enhanced Super-Resoluton System wth Improved Image Regstraton, Automatc Image Selecton, and Image Enhancement Yu-Chuan Kuo ( ), Chen-Yu Chen ( ), and Chou-Shann Fuh ( ) Department of Computer Scence

More information

Who are you with and Where are you going?

Who are you with and Where are you going? Who are you wth and Where are you gong? Kota Yamaguch Alexander C. Berg Lus E. Ortz Tamara L. Berg Stony Brook Unversty Stony Brook Unversty, NY 11794, USA {kyamagu, aberg, leortz, tlberg}@cs.stonybrook.edu

More information

Abstract. Clustering ensembles have emerged as a powerful method for improving both the

Abstract. Clustering ensembles have emerged as a powerful method for improving both the Clusterng Ensembles: {topchyal, Models jan, of punch}@cse.msu.edu Consensus and Weak Parttons * Alexander Topchy, Anl K. Jan, and Wllam Punch Department of Computer Scence and Engneerng, Mchgan State Unversty

More information

SIMPLE LINEAR CORRELATION

SIMPLE LINEAR CORRELATION SIMPLE LINEAR CORRELATION Smple lnear correlaton s a measure of the degree to whch two varables vary together, or a measure of the ntensty of the assocaton between two varables. Correlaton often s abused.

More information

Calculating the high frequency transmission line parameters of power cables

Calculating the high frequency transmission line parameters of power cables < ' Calculatng the hgh frequency transmsson lne parameters of power cables Authors: Dr. John Dcknson, Laboratory Servces Manager, N 0 RW E B Communcatons Mr. Peter J. Ncholson, Project Assgnment Manager,

More information

A novel Method for Data Mining and Classification based on

A novel Method for Data Mining and Classification based on A novel Method for Data Mnng and Classfcaton based on Ensemble Learnng 1 1, Frst Author Nejang Normal Unversty;Schuan Nejang 641112,Chna, E-mal: lhan-gege@126.com Abstract Data mnng has been attached great

More information

Enriching the Knowledge Sources Used in a Maximum Entropy Part-of-Speech Tagger

Enriching the Knowledge Sources Used in a Maximum Entropy Part-of-Speech Tagger Enrchng the Knowledge Sources Used n a Maxmum Entropy Part-of-Speech Tagger Krstna Toutanova Dept of Computer Scence Gates Bldg 4A, 353 Serra Mall Stanford, CA 94305 9040, USA krstna@cs.stanford.edu Chrstopher

More information

Course outline. Financial Time Series Analysis. Overview. Data analysis. Predictive signal. Trading strategy

Course outline. Financial Time Series Analysis. Overview. Data analysis. Predictive signal. Trading strategy Fnancal Tme Seres Analyss Patrck McSharry patrck@mcsharry.net www.mcsharry.net Trnty Term 2014 Mathematcal Insttute Unversty of Oxford Course outlne 1. Data analyss, probablty, correlatons, vsualsaton

More information

Support vector domain description

Support vector domain description Pattern Recognton Letters 20 (1999) 1191±1199 www.elsever.nl/locate/patrec Support vector doman descrpton Davd M.J. Tax *,1, Robert P.W. Dun Pattern Recognton Group, Faculty of Appled Scence, Delft Unversty

More information

Performance Analysis and Coding Strategy of ECOC SVMs

Performance Analysis and Coding Strategy of ECOC SVMs Internatonal Journal of Grd and Dstrbuted Computng Vol.7, No. (04), pp.67-76 http://dx.do.org/0.457/jgdc.04.7..07 Performance Analyss and Codng Strategy of ECOC SVMs Zhgang Yan, and Yuanxuan Yang, School

More information

Cloud-based Social Application Deployment using Local Processing and Global Distribution

Cloud-based Social Application Deployment using Local Processing and Global Distribution Cloud-based Socal Applcaton Deployment usng Local Processng and Global Dstrbuton Zh Wang *, Baochun L, Lfeng Sun *, and Shqang Yang * * Bejng Key Laboratory of Networked Multmeda Department of Computer

More information

L10: Linear discriminants analysis

L10: Linear discriminants analysis L0: Lnear dscrmnants analyss Lnear dscrmnant analyss, two classes Lnear dscrmnant analyss, C classes LDA vs. PCA Lmtatons of LDA Varants of LDA Other dmensonalty reducton methods CSCE 666 Pattern Analyss

More information

Brigid Mullany, Ph.D University of North Carolina, Charlotte

Brigid Mullany, Ph.D University of North Carolina, Charlotte Evaluaton And Comparson Of The Dfferent Standards Used To Defne The Postonal Accuracy And Repeatablty Of Numercally Controlled Machnng Center Axes Brgd Mullany, Ph.D Unversty of North Carolna, Charlotte

More information

A Novel Methodology of Working Capital Management for Large. Public Constructions by Using Fuzzy S-curve Regression

A Novel Methodology of Working Capital Management for Large. Public Constructions by Using Fuzzy S-curve Regression Novel Methodology of Workng Captal Management for Large Publc Constructons by Usng Fuzzy S-curve Regresson Cheng-Wu Chen, Morrs H. L. Wang and Tng-Ya Hseh Department of Cvl Engneerng, Natonal Central Unversty,

More information

+ + + - - This circuit than can be reduced to a planar circuit

+ + + - - This circuit than can be reduced to a planar circuit MeshCurrent Method The meshcurrent s analog of the nodeoltage method. We sole for a new set of arables, mesh currents, that automatcally satsfy KCLs. As such, meshcurrent method reduces crcut soluton to

More information

Bayesian Network Based Causal Relationship Identification and Funding Success Prediction in P2P Lending

Bayesian Network Based Causal Relationship Identification and Funding Success Prediction in P2P Lending Proceedngs of 2012 4th Internatonal Conference on Machne Learnng and Computng IPCSIT vol. 25 (2012) (2012) IACSIT Press, Sngapore Bayesan Network Based Causal Relatonshp Identfcaton and Fundng Success

More information

A hybrid global optimization algorithm based on parallel chaos optimization and outlook algorithm

A hybrid global optimization algorithm based on parallel chaos optimization and outlook algorithm Avalable onlne www.ocpr.com Journal of Chemcal and Pharmaceutcal Research, 2014, 6(7):1884-1889 Research Artcle ISSN : 0975-7384 CODEN(USA) : JCPRC5 A hybrd global optmzaton algorthm based on parallel

More information

Joint Scheduling of Processing and Shuffle Phases in MapReduce Systems

Joint Scheduling of Processing and Shuffle Phases in MapReduce Systems Jont Schedulng of Processng and Shuffle Phases n MapReduce Systems Fangfe Chen, Mural Kodalam, T. V. Lakshman Department of Computer Scence and Engneerng, The Penn State Unversty Bell Laboratores, Alcatel-Lucent

More information

AN APPOINTMENT ORDER OUTPATIENT SCHEDULING SYSTEM THAT IMPROVES OUTPATIENT EXPERIENCE

AN APPOINTMENT ORDER OUTPATIENT SCHEDULING SYSTEM THAT IMPROVES OUTPATIENT EXPERIENCE AN APPOINTMENT ORDER OUTPATIENT SCHEDULING SYSTEM THAT IMPROVES OUTPATIENT EXPERIENCE Yu-L Huang Industral Engneerng Department New Mexco State Unversty Las Cruces, New Mexco 88003, U.S.A. Abstract Patent

More information

Everybody needs somebody: Modeling social and grouping behavior on a linear programming multiple people tracker

Everybody needs somebody: Modeling social and grouping behavior on a linear programming multiple people tracker Everybody needs somebody: Modelng socal and groupng behavor on a lnear programmng multple people tracker Laura Leal-Taxe, Gerard Pons-Moll and Bodo Rosenhahn Insttute for Informaton Processng (TNT) Lebnz

More information

A Secure Password-Authenticated Key Agreement Using Smart Cards

A Secure Password-Authenticated Key Agreement Using Smart Cards A Secure Password-Authentcated Key Agreement Usng Smart Cards Ka Chan 1, Wen-Chung Kuo 2 and Jn-Chou Cheng 3 1 Department of Computer and Informaton Scence, R.O.C. Mltary Academy, Kaohsung 83059, Tawan,

More information

Logistic Regression. Steve Kroon

Logistic Regression. Steve Kroon Logstc Regresson Steve Kroon Course notes sectons: 24.3-24.4 Dsclamer: these notes do not explctly ndcate whether values are vectors or scalars, but expects the reader to dscern ths from the context. Scenaro

More information

A Multi-Camera System on PC-Cluster for Real-time 3-D Tracking

A Multi-Camera System on PC-Cluster for Real-time 3-D Tracking The 23 rd Conference of the Mechancal Engneerng Network of Thaland November 4 7, 2009, Chang Ma A Mult-Camera System on PC-Cluster for Real-tme 3-D Trackng Vboon Sangveraphunsr*, Krtsana Uttamang, and

More information

Staff Paper. Farm Savings Accounts: Examining Income Variability, Eligibility, and Benefits. Brent Gloy, Eddy LaDue, and Charles Cuykendall

Staff Paper. Farm Savings Accounts: Examining Income Variability, Eligibility, and Benefits. Brent Gloy, Eddy LaDue, and Charles Cuykendall SP 2005-02 August 2005 Staff Paper Department of Appled Economcs and Management Cornell Unversty, Ithaca, New York 14853-7801 USA Farm Savngs Accounts: Examnng Income Varablty, Elgblty, and Benefts Brent

More information

Open Access A Load Balancing Strategy with Bandwidth Constraint in Cloud Computing. Jing Deng 1,*, Ping Guo 2, Qi Li 3, Haizhu Chen 1

Open Access A Load Balancing Strategy with Bandwidth Constraint in Cloud Computing. Jing Deng 1,*, Ping Guo 2, Qi Li 3, Haizhu Chen 1 Send Orders for Reprnts to reprnts@benthamscence.ae The Open Cybernetcs & Systemcs Journal, 2014, 8, 115-121 115 Open Access A Load Balancng Strategy wth Bandwdth Constrant n Cloud Computng Jng Deng 1,*,

More information

How To Calculate The Accountng Perod Of Nequalty

How To Calculate The Accountng Perod Of Nequalty Inequalty and The Accountng Perod Quentn Wodon and Shlomo Ytzha World Ban and Hebrew Unversty September Abstract Income nequalty typcally declnes wth the length of tme taen nto account for measurement.

More information

THE DISTRIBUTION OF LOAN PORTFOLIO VALUE * Oldrich Alfons Vasicek

THE DISTRIBUTION OF LOAN PORTFOLIO VALUE * Oldrich Alfons Vasicek HE DISRIBUION OF LOAN PORFOLIO VALUE * Oldrch Alfons Vascek he amount of captal necessary to support a portfolo of debt securtes depends on the probablty dstrbuton of the portfolo loss. Consder a portfolo

More information

An Evaluation of the Extended Logistic, Simple Logistic, and Gompertz Models for Forecasting Short Lifecycle Products and Services

An Evaluation of the Extended Logistic, Simple Logistic, and Gompertz Models for Forecasting Short Lifecycle Products and Services An Evaluaton of the Extended Logstc, Smple Logstc, and Gompertz Models for Forecastng Short Lfecycle Products and Servces Charles V. Trappey a,1, Hsn-yng Wu b a Professor (Management Scence), Natonal Chao

More information

A Multi-mode Image Tracking System Based on Distributed Fusion

A Multi-mode Image Tracking System Based on Distributed Fusion A Mult-mode Image Tracng System Based on Dstrbuted Fuson Ln zheng Chongzhao Han Dongguang Zuo Hongsen Yan School of Electroncs & nformaton engneerng, X an Jaotong Unversty X an, Shaanx, Chna Lnzheng@malst.xjtu.edu.cn

More information

Single and multiple stage classifiers implementing logistic discrimination

Single and multiple stage classifiers implementing logistic discrimination Sngle and multple stage classfers mplementng logstc dscrmnaton Hélo Radke Bttencourt 1 Dens Alter de Olvera Moraes 2 Vctor Haertel 2 1 Pontfíca Unversdade Católca do Ro Grande do Sul - PUCRS Av. Ipranga,

More information

Enterprise Master Patient Index

Enterprise Master Patient Index Enterprse Master Patent Index Healthcare data are captured n many dfferent settngs such as hosptals, clncs, labs, and physcan offces. Accordng to a report by the CDC, patents n the Unted States made an

More information

Searching for Interacting Features for Spam Filtering

Searching for Interacting Features for Spam Filtering Searchng for Interactng Features for Spam Flterng Chuanlang Chen 1, Yun-Chao Gong 2, Rongfang Be 1,, and X. Z. Gao 3 1 Department of Computer Scence, Bejng Normal Unversty, Bejng 100875, Chna 2 Software

More information

GRAVITY DATA VALIDATION AND OUTLIER DETECTION USING L 1 -NORM

GRAVITY DATA VALIDATION AND OUTLIER DETECTION USING L 1 -NORM GRAVITY DATA VALIDATION AND OUTLIER DETECTION USING L 1 -NORM BARRIOT Jean-Perre, SARRAILH Mchel BGI/CNES 18.av.E.Beln 31401 TOULOUSE Cedex 4 (France) Emal: jean-perre.barrot@cnes.fr 1/Introducton The

More information

Hallucinating Multiple Occluded CCTV Face Images of Different Resolutions

Hallucinating Multiple Occluded CCTV Face Images of Different Resolutions In Proc. IEEE Internatonal Conference on Advanced Vdeo and Sgnal based Survellance (AVSS 05), September 2005 Hallucnatng Multple Occluded CCTV Face Images of Dfferent Resolutons Ku Ja Shaogang Gong Computer

More information

"Research Note" APPLICATION OF CHARGE SIMULATION METHOD TO ELECTRIC FIELD CALCULATION IN THE POWER CABLES *

Research Note APPLICATION OF CHARGE SIMULATION METHOD TO ELECTRIC FIELD CALCULATION IN THE POWER CABLES * Iranan Journal of Scence & Technology, Transacton B, Engneerng, ol. 30, No. B6, 789-794 rnted n The Islamc Republc of Iran, 006 Shraz Unversty "Research Note" ALICATION OF CHARGE SIMULATION METHOD TO ELECTRIC

More information

Damage detection in composite laminates using coin-tap method

Damage detection in composite laminates using coin-tap method Damage detecton n composte lamnates usng con-tap method S.J. Km Korea Aerospace Research Insttute, 45 Eoeun-Dong, Youseong-Gu, 35-333 Daejeon, Republc of Korea yaeln@kar.re.kr 45 The con-tap test has the

More information

A DATA MINING APPLICATION IN A STUDENT DATABASE

A DATA MINING APPLICATION IN A STUDENT DATABASE JOURNAL OF AERONAUTICS AND SPACE TECHNOLOGIES JULY 005 VOLUME NUMBER (53-57) A DATA MINING APPLICATION IN A STUDENT DATABASE Şenol Zafer ERDOĞAN Maltepe Ünversty Faculty of Engneerng Büyükbakkalköy-Istanbul

More information

Frequency Selective IQ Phase and IQ Amplitude Imbalance Adjustments for OFDM Direct Conversion Transmitters

Frequency Selective IQ Phase and IQ Amplitude Imbalance Adjustments for OFDM Direct Conversion Transmitters Frequency Selectve IQ Phase and IQ Ampltude Imbalance Adjustments for OFDM Drect Converson ransmtters Edmund Coersmeer, Ernst Zelnsk Noka, Meesmannstrasse 103, 44807 Bochum, Germany edmund.coersmeer@noka.com,

More information

ECE544NA Final Project: Robust Machine Learning Hardware via Classifier Ensemble

ECE544NA Final Project: Robust Machine Learning Hardware via Classifier Ensemble 1 ECE544NA Fnal Project: Robust Machne Learnng Hardware va Classfer Ensemble Sa Zhang, szhang12@llnos.edu Dept. of Electr. & Comput. Eng., Unv. of Illnos at Urbana-Champagn, Urbana, IL, USA Abstract In

More information

Document Clustering Analysis Based on Hybrid PSO+K-means Algorithm

Document Clustering Analysis Based on Hybrid PSO+K-means Algorithm Document Clusterng Analyss Based on Hybrd PSO+K-means Algorthm Xaohu Cu, Thomas E. Potok Appled Software Engneerng Research Group, Computatonal Scences and Engneerng Dvson, Oak Rdge Natonal Laboratory,

More information

v a 1 b 1 i, a 2 b 2 i,..., a n b n i.

v a 1 b 1 i, a 2 b 2 i,..., a n b n i. SECTION 8.4 COMPLEX VECTOR SPACES AND INNER PRODUCTS 455 8.4 COMPLEX VECTOR SPACES AND INNER PRODUCTS All the vector spaces we have studed thus far n the text are real vector spaces snce the scalars are

More information

Distributed Multi-Target Tracking In A Self-Configuring Camera Network

Distributed Multi-Target Tracking In A Self-Configuring Camera Network Dstrbuted Mult-Target Trackng In A Self-Confgurng Camera Network Crstan Soto, B Song, Amt K. Roy-Chowdhury Department of Electrcal Engneerng Unversty of Calforna, Rversde {cwlder,bsong,amtrc}@ee.ucr.edu

More information

Conversion between the vector and raster data structures using Fuzzy Geographical Entities

Conversion between the vector and raster data structures using Fuzzy Geographical Entities Converson between the vector and raster data structures usng Fuzzy Geographcal Enttes Cdála Fonte Department of Mathematcs Faculty of Scences and Technology Unversty of Combra, Apartado 38, 3 454 Combra,

More information

PRACTICE 1: MUTUAL FUNDS EVALUATION USING MATLAB.

PRACTICE 1: MUTUAL FUNDS EVALUATION USING MATLAB. PRACTICE 1: MUTUAL FUNDS EVALUATION USING MATLAB. INDEX 1. Load data usng the Edtor wndow and m-fle 2. Learnng to save results from the Edtor wndow. 3. Computng the Sharpe Rato 4. Obtanng the Treynor Rato

More information

Data Broadcast on a Multi-System Heterogeneous Overlayed Wireless Network *

Data Broadcast on a Multi-System Heterogeneous Overlayed Wireless Network * JOURNAL OF INFORMATION SCIENCE AND ENGINEERING 24, 819-840 (2008) Data Broadcast on a Mult-System Heterogeneous Overlayed Wreless Network * Department of Computer Scence Natonal Chao Tung Unversty Hsnchu,

More information

APPLICATION OF PROBE DATA COLLECTED VIA INFRARED BEACONS TO TRAFFIC MANEGEMENT

APPLICATION OF PROBE DATA COLLECTED VIA INFRARED BEACONS TO TRAFFIC MANEGEMENT APPLICATION OF PROBE DATA COLLECTED VIA INFRARED BEACONS TO TRAFFIC MANEGEMENT Toshhko Oda (1), Kochro Iwaoka (2) (1), (2) Infrastructure Systems Busness Unt, Panasonc System Networks Co., Ltd. Saedo-cho

More information

Transition Matrix Models of Consumer Credit Ratings

Transition Matrix Models of Consumer Credit Ratings Transton Matrx Models of Consumer Credt Ratngs Abstract Although the corporate credt rsk lterature has many studes modellng the change n the credt rsk of corporate bonds over tme, there s far less analyss

More information

Risk-based Fatigue Estimate of Deep Water Risers -- Course Project for EM388F: Fracture Mechanics, Spring 2008

Risk-based Fatigue Estimate of Deep Water Risers -- Course Project for EM388F: Fracture Mechanics, Spring 2008 Rsk-based Fatgue Estmate of Deep Water Rsers -- Course Project for EM388F: Fracture Mechancs, Sprng 2008 Chen Sh Department of Cvl, Archtectural, and Envronmental Engneerng The Unversty of Texas at Austn

More information

STANDING WAVE TUBE TECHNIQUES FOR MEASURING THE NORMAL INCIDENCE ABSORPTION COEFFICIENT: COMPARISON OF DIFFERENT EXPERIMENTAL SETUPS.

STANDING WAVE TUBE TECHNIQUES FOR MEASURING THE NORMAL INCIDENCE ABSORPTION COEFFICIENT: COMPARISON OF DIFFERENT EXPERIMENTAL SETUPS. STADIG WAVE TUBE TECHIQUES FOR MEASURIG THE ORMAL ICIDECE ABSORPTIO COEFFICIET: COMPARISO OF DIFFERET EXPERIMETAL SETUPS. Angelo Farna (*), Patrzo Faust (**) (*) Dpart. d Ing. Industrale, Unverstà d Parma,

More information

Feature selection for intrusion detection. Slobodan Petrović NISlab, Gjøvik University College

Feature selection for intrusion detection. Slobodan Petrović NISlab, Gjøvik University College Feature selecton for ntruson detecton Slobodan Petrovć NISlab, Gjøvk Unversty College Contents The feature selecton problem Intruson detecton Traffc features relevant for IDS The CFS measure The mrmr measure

More information

Human Tracking by Fast Mean Shift Mode Seeking

Human Tracking by Fast Mean Shift Mode Seeking JOURAL OF MULTIMEDIA, VOL. 1, O. 1, APRIL 2006 1 Human Trackng by Fast Mean Shft Mode Seekng [10 font sze blank 1] [10 font sze blank 2] C. Belezna Advanced Computer Vson GmbH - ACV, Venna, Austra Emal:

More information

Web Spam Detection Using Machine Learning in Specific Domain Features

Web Spam Detection Using Machine Learning in Specific Domain Features Journal of Informaton Assurance and Securty 3 (2008) 220-229 Web Spam Detecton Usng Machne Learnng n Specfc Doman Features Hassan Najadat 1, Ismal Hmed 2 Department of Computer Informaton Systems Faculty

More information

Tracking with Non-Linear Dynamic Models

Tracking with Non-Linear Dynamic Models CHAPTER 2 Trackng wth Non-Lnear Dynamc Models In a lnear dynamc model wth lnear measurements, there s always only one peak n the posteror; very small non-lneartes n dynamc models can lead to a substantal

More information

Enabling P2P One-view Multi-party Video Conferencing

Enabling P2P One-view Multi-party Video Conferencing Enablng P2P One-vew Mult-party Vdeo Conferencng Yongxang Zhao, Yong Lu, Changja Chen, and JanYn Zhang Abstract Mult-Party Vdeo Conferencng (MPVC) facltates realtme group nteracton between users. Whle P2P

More information

CHOLESTEROL REFERENCE METHOD LABORATORY NETWORK. Sample Stability Protocol

CHOLESTEROL REFERENCE METHOD LABORATORY NETWORK. Sample Stability Protocol CHOLESTEROL REFERENCE METHOD LABORATORY NETWORK Sample Stablty Protocol Background The Cholesterol Reference Method Laboratory Network (CRMLN) developed certfcaton protocols for total cholesterol, HDL

More information

Quantization Effects in Digital Filters

Quantization Effects in Digital Filters Quantzaton Effects n Dgtal Flters Dstrbuton of Truncaton Errors In two's complement representaton an exact number would have nfntely many bts (n general). When we lmt the number of bts to some fnte value

More information