Download - Metadata 0504002 02_020

Transcript
  • 1. Intelligent Surveillance for Parking Lots: People andCars Detection [email protected] [email protected] [email protected] surveillance for parking lots. This method includes fast background subtraction, motion object behavior analysis, and motion object classification. Thebackground subtraction method need not background model in advance. A Fibonacci Series based background updating is proposed to update the background model. The fast background subtraction is used to detect motion object under the fixed or movable camera. In the motion object behavior analysis stage, a motion analytical table is proposed to record the movable behavior. The motion analytical table is used to divide the motion object into five behaviors: stop, move, move and stop, stop and move, changing the camera position. These five behaviors are used to determine the speed of the background updating. In the motion object 13 320x240 classification phase, humans skin, hair, eyes, mouth, and the ratio of the motion object are used to identify the 4 Frames motion object as people or car. The identification method is rule-based. The experiment shows that the fastThis paper proposes people and car background subtraction is effectivedetection method in intelligent 1

2. which can deal 320x240 frame size with 13 fps. Especially the speed of thebackground updating is the fastest than other methods. The speed of the background updating only uses 4Frames. (frame differencing)[1][4][6] Background Subtraction,Fibonacci number, background updating, motion analytical table optical flow [5][6] vector field (Pixel) [1](backgroundsubtraction)[1][2][3] 2 3. 1.1 1.1 ()BackgroundForeground CCD Camera 3 4. [8] T =+ 0.2 2.8 T RGB RGB YCbCr Y (Pepper and Salt) (Median Filter) Thresholding T T (Dilation Operation)T (Erosion Operation)T T (Closing) Otsu[7] Otsu Connected component [8] T [8] T T = (Luminous) 4 5. 1 2 3.1 3 412345 67 8 x y 3.1 (+, +) (-,-) 0: 12 3 5 6. PB +1 ( x, y )+ = min ( , PF ( x, y ) PB ( x, y ) ) Sign (PF ( x, y ) PB ( x, y ) ) + 1 ifx 0Sign ( x) = 1 ifx < 0 (3.1) PF(x, y) [9]Frame(x,y)(Ghost)PB(x, y) (x,y) 6 7. 45 0 255 12341234 0 255 (Fibonacci Series) Fib(1) Fib(2) Fib(5) Fib(7) Fib(9) Fib(8) CPU P42.4G/512MB RAM BorlandC++ Builder 6 1237 8. () 5.1 5.1(a)(a) (b)5.1(b) 5.1(a),5.1(c), 5.1(e),5.1(g) 5.1(i)(c) (d)( 5.1(a)) =55 (= Fib(9)) 4 Frames 5.1(i)(e) (f) [10] 60~120 Frames [11] 20 () (g) (h) 5.2 5.2(a)(Frame #84)(i)(j) 5.2(b)(Frame #137)5.1 5.2(c)(Frame #197) Logitech QuickCam Pro 5.2(h)(Frame 320240670)12 345 (6) () 8 9. (a)#84(b)#137(a)#29(b)#70(c)#197(d)#443(c)#140 (d)#284(e)#564(f)#596(e)#68(f)#527(g)#644(h)#670(g)#544 (h)#7235.2 5.3 Frame#29 Frame #70 ( 5.3(b)) (i)#927(j)#10465.3 Frame #140 5.3(c) Frame #284 (5.3(d)) Frame 544( 5.3(g)) Frame #290 ( 5.3(e)) Frame #527 ( 5.3(f)) 9 10. (a)#5(b)#16 (a)#19(b)#39 (c)#32(d)#44 (c)#45 (d)#575.5 () (e)#52(f)#65 5.5 ( 5.5(a))( 5.5(c)) Frame #57( 5.5(d))(g)#75(h)#83 5.4 5.3(g) 5.3(h) 5.3(j)() 5.4 Frame #5 Frame #16 () 5.4(e) 320 x 240 13 [11][12][13] 11 10 11. 5.1 13 frames T FFR CECC (1),(2) , Video 130 39013 0 100%(3) ,(4) Video 260 76512.757 99.08% Video 390115512.83 13 98.87%Gaussian[12]X X 11.25X X This work was supported by the Ministry of Economic, R.O.C., underMixture [11]X X11* X X Grants MOEA-94-EC-17-A-02-S1-032 and National Science Council, R.O.C.,BayesianX X 11 X X under Grants NSC 94-2213-E-133-001-.T: Time (sec.), F: Frames, FR: Frame Rate, CE: Classification Error, CC: Classification Correct. *: image size 160x120 [1] Collins, Lipton, Kanade, Fujiyoshi,( 5.1)Duggins, Tsin, Tolliver, Enomoto, and 4Hasegawa, A System for Video Surveillance and Monitoring: VSAM Final Report, Technical report[10] 60 120CMU-RI-TR-00-12, Robotics Institute, Carnegie Mellon University, May, 2000. [2] I. Haritaoglu, Larry S. Davis, and D. Harwood, W4 who? when? where? what? a real time system for detecting andtracking people, In FGR98, 1998. [3] C. Wren, A. Azarbayejani, T. Darrell, and Alex Pentland, Pfinder: Real-time tracking of the human body, IEEE Transactions on Pattern Analysis and Machine Intelligence, vol. 19, no. 7, pp780785, 1997. [4] C. Anderson, Peter Burt, and G. van der Wal, Change detection and tracking using pyramid transformation techniques, in Proceedings of SPIE - Intelligent Robotsand Computer Vision, vol. 579, pp 7278, 1985. [5] J. Barron, D. Fleet, and S. Beauchemin, Performance of optical flow techniques, International Journal of Computer Vision, vol. 12, no. 1, pp4277, 1994. [6] Ying-li Tian and Arun Hampapur, Robust 4 Salient Motion Detection with Complexframes Background for Real-time Video Surveillance, in IEEE Computer Society11 12. Workshop on Motion and Video Computing,Luo, Fast and Robust Background Breckenridge, Colorado, January 5 and 6, Updating for Real-time Traffic Surveillance 2005.and Monitoring, IEEE Computer Society[7]N. Otsu, A Thresholding Selection MethodConference on Computer Vision and Pattern from Gray-level Histogram, IEEE Recognition, vol. 3, pp. 55 60, June 2005. Transactions System Man Cybernet, vol. 9, [11]Christ Stauffer, W.E.L Grimson, Adaptive pp. 62 66, 1979. Background Mixture Models for Real-time[8]Feng-Sheng Chen, Chih-Ming Fu, Tracking, IEEE Computer Society on Chung-Lin Huang , Hand GestureComputer Vision and Pattern Recognition, Recognition A Real-Time Tracking and vol. 2, June 1999. Hidden Markov Models, Image and Vision [12] Tiehan Lv, Burak Ozer, Wayne Wolf, A Computing , vol. 21, no. 8, pp. 745 758, Real-Time Background Subtraction Method March 2003 with Camera Motion Compensation, IEEE[9]Rita Cucchiara, Massimo Piccardi and International Conference on Multimedia Andrea Prati, Detecting Moving Objects, and Expo, vol. 1, pp. 331 331, June 2004. Ghost, and Shadows in Video Streams, [13] Yaser Sheikh , Mubarak Shah, Bayesian IEEE Transaction on Pattern Analysis and Object Detection in Dynamic Scenes, Machine Intelligence, vol. 25, no. 10, IEEE Compute Society Conference on October 2003.Computer Vision and Pattern Recognition,[10] Suchendra M. Bhandarkar and Xingzhivol. 1, pp. 74 79, June 2005.12