COMPUTER VISION SYSTEMS
COMPUTER VISION SYSTEMS CAP 6411
University of Central Florida
Popular in Course
Popular in System Engineering
This 11 page Class Notes was uploaded by Khalil Conroy on Thursday October 22, 2015. The Class Notes belongs to CAP 6411 at University of Central Florida taught by Staff in Fall. Since its upload, it has received 21 views. For similar materials see /class/227217/cap-6411-university-of-central-florida in System Engineering at University of Central Florida.
Reviews for COMPUTER VISION SYSTEMS
Report this Material
What is Karma?
Karma is the currency of StudySoup.
You can buy or earn more Karma at anytime and redeem it for class notes, study guides, flashcards, and more!
Date Created: 10/22/15
University of Central Florida Probab Istic Tracking in a Metric Space Kentaro Toyama and Andrew Blake Microsoft Research Presentation prepared by Linus Luotsinen Outline I Introduction I Modelling of images and observations I Pattern theoretic tracking I Learning Leam mixture centers exemplars Leam kemel parameters observational likeli a hood Le m dynamic model transition probabllltles I Practical tracking I Results Human motion using curve based exemplars Mouth using exemplars from raw 39ma I Conclusions 6 Univurbily of Central Florida University of Central Florida Introduction Metric Mixture M2 Combine exemplars in metric space with probabilistic treatments Models easily created directly from training set Dynamic model to deal with occlusion Problems with other probabilistic approaches Complex models Training required for each object to be tracked Difficult to fully automate 6 L11 Iivursity of Conlrul Florida Pattern Theoretic Tracking Notation Test images Z zlzzzT Train images 2 zzz Class is defined by a set of exemplars X Epic 12K Geometrical transformation T a a e A known in advance is Fill Pattern theoretic tracking 2 z T ax State vector defined by OLA Q l b 6 University of Central Florida University of Central Florida Metric Functions 0 True metric function 1pa b gt 0 Va b 9 9 9 All constraints 0 Distance function 2 Pagtb 01 a b Without 3 and 4 5 5 9 9 a b C 6 Universin of Central Florida Modelling of Images and Observations Patches Image subregion Shuffle distance function Curves Edge maps Chamfer distance function Distance to the nearest pixel in the binary images See next slide 6 University of Central Florida University of Central Florida Probabilistic Modelling of Images and Observations Curves with Chamfer distance if iginal i Feature image image I l pchamfer1 E26110 l lnivursily ul39 Zunlral Florida V I A M D15tance Exemplar T image 7 dI Not a true metric 3p1T MT 4 MA B MB 1 CHMLC Pattern Theoretic Tracking 3 E gquot zmTax Observation Geometrical Transform Translation Affine Projective e Univursily of Ccmral Florida Exemplar from a training set Intensity images Feature images patches edges corners University of Central Florida Pattern Theoretic Tracking ax amp e Univursily of Central Florida Pattern Theoretic Tracking Learning 1 AsetofKExemplars l l f l lll 2 p2 l X Distribution of observations Xak around T afk 7 Likelihood in amquot H 3 7Xt l X H Prior about dependency between states 7 Dynamics 6 Univurbily of Central Florida quot1 University of Central Florida Learning Mixture Centers GoalgivenMimageszm it it 0 o f t it it M ndKexemplars f G 9 g 0 g p 3 g A l 9 4 l l 1 I ll l A lF1nd central exemplar Q Q 0 D l 1 A A 20 argminmaxpz z 0 o D 0 0 ll ll A a z z39 7 0 F U Q ll ll ll 0 0 0 ll l A ll 2 Align otherimagesto 20 9 f 9 ll 8 Q ll 43 g 1 OGQPMAXatQAA amZargngnpTa Zmazo 0 I R Q Q A A 1 V maa mn m1M Universin of Central Florida Learning Mixture Cente 3 Find K distinct exemplars Q 9 0 f A pfk13 fk H pa 4 Cluster the rest by minimal distance km argmkin pm 5 Find new representatives N xk arg mxm man 0x x 6 Universily uf Ccnlml Florida University of Central Florida Learning Kernel Parameters 1 Using a validation set nd distances between images and their exemplars p Z 7 To xk 2 Approx distances as chisquare N 022 to nda andd 3 Then the observation likelihood is l 2039 2 pzX ocexp tp Z oc ad 2 6 Universin of Central Florida Learning Dynamics 0 Learn a Markov matrix Mfor pklk1 by histogramming transitions 0 Run a first order autoregressive process ARP for Mariam with coef cients calculated using the Yule Walker algorithm 11 p 6 Universily ul Central Florida University of Central Flonda Practical Tracking 0 Forward algorithm PXEPX Zperwzr we 2 j m l XgtpltX XmeXka 94 04H 0 Results are chosen by X argmaxpX t 1 2 6 l lnivursily oi Icnlral Florida Results 0 Tracking human motion Based on contour edges Dynamics learned on 5 sequences of 100 frames each Exemplars Same person motion not seen in training sequence a U1 ivursily 0i Ccmral Florida University of Central Flonda Results 0 Tracking human motion Based on contour edges Dynamics learned on 5 sequences of 100 frames each Different Person Different person with occlusion power of dynamic model 6 University of Ccnlral Florida Results 0 Tracking person s mouth motion Based on raw pixel values Training sequence was 210 frames captured at 30Hz Exemplar set was 30 K30 Left image show test sequence 0 Right image show maximum a posteriori Using L2 distance Using shuffle distance 6 Universin 0i Contra Florida University of Central Florida Results I Tracking ballerina Larger exemplar sets K300 6 Universin of Central Florida Conclusions I Metric Mixture M2 Mode Easier to fully automate learning I Generality trics can be chosen without signi cant restrictions I Temporal fusion of information for occlusion recovery Bayesian networks 6 Universily of Central Florida University of Central Florida References I 1 Kentaro Toyama Andrew Blake Probabilistic Tracking with Exemplars in a Metric Space International Jour al of Computer Vision Volume 48 Is ue Marr Prize Special Issue Pages 9 19 2002 ISSN092056 I 2 Jongwoo Lim CSE 252C Selected Topics in Visio amp Learning http wwwcseucsdeduIclassesfaD2cse252c I 3 Eli Schechtman and Neer Saad Advanced topics in computer and human vision rsehtm 6 University of Central Florida
Are you sure you want to buy this material for
You're already Subscribed!
Looks like you've already subscribed to StudySoup, you won't need to purchase another subscription to get this material. To access this material simply click 'View Full Document'