Toward Real-Time Extraction of Pedestrian ... - Keio University

ht.sfc.keio.ac.jp

Toward Real-Time Extraction of Pedestrian ... - Keio University

Toward Real-Time Extraction of Pedestrian

Contexts with Sterao Camera

Kei Suzuki 1 , Kazunori Takashio 2 , Hideyuki Tokuda 1,2 , Masaki Wada 3 , Yusuke Matsuki 3 , Kazunori Umeda 3

1 Graduate School of Media and Governance, Keio University,

2 Faculty of Environmental Information, Keio University and

3 Dept. Precision Mecanics, Faculty of Science and Engineering Chuo Univ. / CREST, JST

Tokyo, Japan

Email: suzuk@ht.sfc.keio.ac.jp, kaz@mkg.sfc.keio.ac.jp

wada@sensor.mech.chuo-u.ac.jp, umeda@mech.chuo-u.ac.jp

Abstract—Our goal is to extract the mood of disquiet on

street corners in real-time with stereo video camera systems.

In the last year’s INSS, we proposed a novel stereo mesurement

algorithm to detect moving people, which was focusing on moving

region in video data. In this paper, we report our prototype of

probabilistic inference engine that can detect contexts of each

individual pedestrian and those of target pedestrian groups or

mobs. Here, we have tried to apply the Bayesian Network Model

to real-time context analysis.

Index Terms—distributed camera system, stereo vision, realtime

context analysis, pedestrian context, bayesian network

model.

I. INTRODUCTION




[1],[2],[3]




3











18times/sec


x,y,z



1.56m







Fig.2

Fig.3(a),(b)

Fig.3(a)

Fig.3(a) Fig.3(b)






















III. EXTRACTION PEDESTRIAN CONTEXTS FROM STERAO

VISION DATA





II.


Fig.1

A. Features of pedestrian contexts

3


(a) With motion detection(proposed method)

Fig. 1: Flow or our stereo algorithm

(b) With motion detection(proposed method)

Fig. 3: An example of our stereo algorithm: range images for

Fig.2

Fig. 2: An example scene

B. Bayesian Network Approach





,

.



.






2








[5],[4]














[7]






Likelihood Weighting

[6]


Fig. 4: Thumble bayesian network model

Fig. 5: Friend group bayesian network model

C. Bayse model of Target pedestrian contexts



.




1) Context of individual pedestrian:


Fig.4


Thumble xs1, ys1,

z1 1 X , Y

, Z xs2, ys2, z2 xs1,

ys1, z1

z



Thumble




2) Context of group or mobs:


Fig.5

Friend

Group vs1, dist1, angle1

2 1

vs2, dist2, angle2

vs1, dist1, angle1 vs3,

dist3, angle3 vs2, dist2, angle2 1





IV. REAL-TIME EXTRACTION OF PEDESTRIAN CONTEXTS



















z






2

1

1








C++GUI Qt

V. EXPERIMENTS




Point


(a) Experimental scene

Fig. 7: Prototype system

(b) Subtraction image

Fig. 6: Outdoor Example scene

Grey Research Bumblebee2f=3.8mm, XGA

320 240



CPUCore2Duo

E6850(2.6GHz 2 core)OSWindows Vista


Fig.6(a),6(b)






Bumblebee2 18fps


Fig.7


7times/sec





58msec








VI. CONCLUSION












ACKNOWLEDGMENT

The authors would like to thank...

REFERENCES

[1] M. Bertozzi and E. Binelli and A. Broggi and M. Del Rose, Stereo Visionbased

approaches for Pedestrian Detection, CVPR ’05: Proceedings of

the 2005 IEEE Computer Society Conference on Computer Vision and

Pattern Recognition (CVPR’05) - Workshops, 2005.

[2] Z. Liang and C. Thorpe, Stereo- and Neural Network-Based Pedestrian

Detection, Proceedings of the IEEE Intelligent Transportation Systems

Conference, Tokyo, Japan, Spring, 1999.

[3] F. Xu and K. Fujimura, Pedestrian detection and tracking with night

vision, In IEEE Intelligent Vehicles Symposium, 2002. 212.

[4] R. E. Neapolitan. Learning Bayesian Networks. Prentice Hall, 2003.

[5] S. Russel and P. Norvig. Artificiall Intelligence Modern Approach Second

Edition. Prentice Hall, 2002.

[6] R. Fung, Chang, and K.-C. Weighting and integrating evidence for

stochasitic simulation in bayesian networks. In Fifth Conference on

Uncertainty in Artificial Intelligence, 1989.

[7] H. Guo and W. Hsu. A survey of algorithms for real-time bayesian

network inference, 2002.

More magazines by this user
Similar magazines