A real-time video surveillance system with human occlusion handling using nonlinear regression
Hele tekst
(2) A REAL-TIME VIDEO SURVEILLANCE SYSTEM WITH HUMAN OCCLUSION HANDLING USING NONLINEAR REGRESSION Jungong Hana , Minwei Fenga , and Peter H.N. de Witha,b a. University of Technology Eindhoven, 5600MB Eindhoven, The Netherlands b CycloMedia Technology, 4180 BB Waardenburg, The Netherlands [email protected] ABSTRACT.
(3)
(4) ! " # "# $ # . . . "# %
(5) ! . $
(6) % . & . ' % . %
(7) ( % ( )*
(8) +* . 1. INTRODUCTION ,
(9) ! ( -% % % #
(10). % $
(11) % ( . ! ( % % / . 0 ( !
(12) 1( 2)3
(13)
(14) ! % ( W 4 2+3
(15) . !
(16). -%
(17) 243
(18)
(19) $ ( . !
(20)
(21) ( $ 253 .
(22) & 6 ( %. . ! ! 7
(23) . !
(24). 978-1-4244-2571-6/08/$25.00 ©2008 IEEE. Fig. 1 - ! . / ! 283 . & ! % . # ! %! . % . real-time !
(25) % % # ! &
(26)
(27) & . "# %! 0 .
(28) . %
(29) & . 9
(30) (
(31) ( % ( 9 . . % . . 2. SYSTEM OVERVIEW # ) % :% % ) Object segmentation. 0 ! 2;3 ! . 305. Authorized licensed use limited to: Eindhoven University of Technology. Downloaded on April 8, 2009 at 09:18 from IEEE Xplore. Restrictions apply.. ICME 2008.
(32)
(33) Occlusion detection. ! . ! " ! ! # ! . $ % ! !
(34) & ' Object tracking. '
(35) ( ! ) Trajectory generation. * &$ ( !
(36) + ! , - .,-/
(37) 0 ! ( & " 1 & (
(38)
(39) ,- &$ 2'+. Fig. 2 %& 81 1 ! (( 1 . 4 U (f |d) $ 9 f & $ F
(40) B : d ( Ci 5 i Vi (f, d) $ # 1 Vi (f, d) =.
(41) ( ./ ( & !
(42) $ 3 $ ! ( $ 4!( 5 $
(43) ( ./ 5 !
(44) ! $ 3 3 ( $ $ $ ! ! ( ( " ! 5 ! ! $ ! ( 3 $ (
(45) 2 ( .
(46) ( . Gaussian. $ $ ! Gaussian ( ! $ ! ! ( ! -$ ! $ ! (( & $ # $ $ $ $3 $ ( $ $ ! ! $ $ 6 * $ & $ $ $ $ 3 # 1. X X | {z } | {z } Vi (f, d) +. i. unaffected by label of j. Vi,j (f, d) .. α1 fi = Ii Ti = 255 α2 fi = Ii Ti = 0 0 ! . ./. ! I
(47) & " $ Ii = F : ! Ii = B Ti ( i $ Canny " Ti = 255 i $ Vi,j (f, d) $ $ 4 ! ; 3 9 1. 3. TEXTURE-CONTROLLED OBJECT SEGMENTATION. U (f |d) =. 8>< >:. Vi,j (f, d) =. 8>< >:. β1 Ii = Ij Ti = 255 β2 Ii = Ij Ti = 0 0 ! . .'/. " α1 = 3 β1 = 1 α2 = β2 = 2 ! $ < $ ( $ $ . 4. OCCLUSION HANDLING USING NONLINEAR REGRESSION
(48) ! $ $ ! $ '=> ! !
(49) > $ " > ! $ & . $ 3 # $ . 75 × 50 / 0 $ ! $ " ! $ 4!( 5 > ! 1 .7/ ! ( $ ! ( ( : ./
(50) > $ & ! #$ model & $ ?! ! ( . smooth ( ! $ $ 5 " ( ( % ! . $ . .7/. (i,j)∈+i. depending on label of j. 306 Authorized licensed use limited to: Eindhoven University of Technology. Downloaded on April 8, 2009 at 09:18 from IEEE Xplore. Restrictions apply..
(51) Table 1. 3
(52) . . Fig. 3 .
(53)
(54) .
(55) . . . . 3. 4). 2 . 5 6. " . 6778. &* &8. 5 9. " . :7 ;8. ;1 &8. 6778 ;/ /8. 5 *. " . 6778. 6778. 6778. 5 /. " . ;7 :8. && 18. &1 :8. 5 1. " . 1/ ;8. && 98. <& :8. 5 :. " . 6778. 6778. 6778. 4.1. Locate the people during the occlusion. . . ! ! ( . . . !
(56) .
(57)
(58) . 2D. . . . ! " .
(59) . y .
(60) . x. # . . y .
(61) .
(62)
(63)
(64) ). *. . {xi , yi }i=m/2...m. . . + ! . F (c, x). μ. c. . . . {xi , yi }i=1...m. . . . xi. yi. . . x . y . i. . . .
(65) . ith. (xi , yi ). . . . . . . . χ2
(66) . . ! m (xi , yi )
(67) F (c, x) !. . χ2 =. : m. (yi − F (c, xi ))2 .. $/(. . F (c, x). . . . c = {c0 , c1 , . . .}. Gaussian. . Gaussian. . !. .
(68) + ! . . ?
(69) . . -. . %*' . . . . )
(70) " . ? . n
(71) n . . .
(72) . .
(73)
(74) . . .
(75) . {ui }i=1...m nth. Tˆn (ui ) > ui i
(76) ) . .
(77) . .
(78) .
(79) .
(80) . 2
(81) . 4.2. Recognize the people during the occlusion. . i=1
(82) !. σ. . >! . .
(83) . - + !. $
(84) ( . . $1(. ! ! .
(85) .
(86) . =..
(87)
(88) .
(89) . ,. " .
(90)
(91) .
(92) . ! $ ( . .
(93) ! . # . Gaussian.
(94) . {xi , yi }i=1...m/2. - !
(95) .
(96) ! . . ! . $ !. . . $ %&' ( . {xi , yi }i=1...m.
(97) + . + . . . . ˆ j (ui ) W. . . . . . . . ! ! 0 ! . F (c, x) = c1 exp(−. ˆ j (ui ), Tˆn (ui )). Cj = arg max ρ(W. . 2. (x − c2 ) (x − c5 ) ) + c4 exp(− ). 2c23 2c26. $1( . !. . ! . . ! . $:(. n. 2. ˆ j (ui ), Tˆn (ui )) ρ(W ) ρ . .
(98) + %*' .
(99)
(100) . + !. c !
(101) ! 2 .
(102) F (c, x) + . 5. EXPERIMENTAL COMPARISONS. ! $ "( ! . xi F (c, xi ) ≥ F (c, xj ) |xj − xi | < !
(103) F (c, xi ) !
(104) ! .
(105) .
(106) .
(107) . > 0. 3 . . 2
(108) . . " . . !.
(109)
(110) . 2. $ . . ! . !
(111). !. 1, 000
(112) ( . !
(113)
(114) . @*7=. # !. + . . "
(115) . " #
(116) . ! !. . " .
(117) . 2 ! . . 70%
(118) # ×9/7 91>
(119) . ! . *97. 307 Authorized licensed use limited to: Eindhoven University of Technology. Downloaded on April 8, 2009 at 09:18 from IEEE Xplore. Restrictions apply..
(120) Fig. 4.
(121)
(122) . . .
(123)
(124)
(125) . .
(126)
(127) . ! "
(128) #$% "&
(129) '(( #)% +
(130)
(131) "
(132) . 70%. >
(133) . *. * . . . "
(134) . 82.7%
(135)
(136) * " &
(137)
(138) 96.4%
(139) "
(140) 93.3% " ,
(141). = & *
(142)
(143)
(144). . *". * "
(145)
(146) . . Fig. 5. ) - & . . 9
(147)
(148)
(149) *
(150) &
(151) . real-time 90%
(152)
(153) . .
(154) . . & - . 7. REFERENCES. - * "
(155)
(156) #/% &
(157) . 0 . #+%. * -". "
(158) A .BBB *. - & . . . +/ 3 #'%.
(159) - . ,
(160).
(161)
(162)
(163) - . occlusion . #$%. . . . < . @!). #)%. * - 6".7 $89 6:. :. , 8.
(164) . : 7. 6. C. D B. @C" . C" . 7 8@? . + . //"++( '(($. E F *@?
(165) ". F. F @. 8 - .
(166) A 6 #3%. 6. C 6. . =. :. .:6. . '4"$+ '((). * @<
(167) & . & A .BBB *. 6. CONCLUSIONS. 6?. . '/ +' '(3/"'(44 '((3 #4%. G. <. , 6 . D. ! ! @. * : B 1' .
(168) . , -
(169)
(170) ,"
(171) = . " .
(172) A .BBB. .
(173) " . - . '''"''3.
(174)
(175) A 6 : .::7 +(3/"+(41 '(($ #5%. ; . <. '+ #1%. 1 " .
(176) <. . " &A .
(177) 7 :
(178) .
(179)
(180) 1) 1' 11$ 34 )5 33 ". . .
(181) A .BBB * 6?. '1 1 15)"133 '(($. - .
(182)
(183)
(184) ". *
(185) ,
(186). <. +//4. " "& 2 '( )1 1() /' . .
(187) . A 6. 1
(188) .
(189) -
(190) . / (3 . .. 6?. . 34("341 +//3.
(191)
(192) ". .
(193)
(194) " & . : ! ? ?9 * < ? 6 @6& . #/%. .
(195) < < @. $ . A .BBB * 4 . 4(/"4$( '(((. 308 Authorized licensed use limited to: Eindhoven University of Technology. Downloaded on April 8, 2009 at 09:18 from IEEE Xplore. Restrictions apply.. 43("434 '((5. " 6?. . '' .
(196)
GERELATEERDE DOCUMENTEN
DOI: 10.1107/S1600536807058850 Document status and date: Published: 01/01/2007 Document Version: Publisher’s PDF, also known as Version of Record includes final page, issue and
We present EnsembleSVM, a free software package con- taining efficient routines to perform ensemble classifi- cation with support vector machine (SVM) base mod- els (Claesen et
As the estimation of model parameters depends to some extent on the quality of the external control spikes, it is advisable to check the quality of the external controls and
f~ Winstknopen: (rente-opbrengst, toename van volume of gewicht door lucht of water- opname etc.). De tot ale uitgaande stroom is gelijk aan de met een factor
Publisher’s PDF, also known as Version of Record (includes final page, issue and volume numbers) Please check the document version of this publication:.. • A submitted manuscript is
Copyright and moral rights for the publications made accessible in the public portal are retained by the authors and/or other copyright owners and it is a condition of
Copyright and moral rights for the publications made accessible in the public portal are retained by the authors and/or other copyright owners and it is a condition of
Vanaf het moment dat er archeologische sporen werden aangetroffen werden deze opgegraven zoals opgelegd door de minimumnormen voor het archeologisch onderzoek en door de