о뷢չ
JOURNAL OF COMPUTER RESEARCH AND DEVELOPMENT
1999ꡡ3610ڡVol.36No.101999



ģ͵˶
СׯԽͦƺ
ժҪڼӾ˶о㷺ӦǰԽԽܵоߵ.ͼе˶иеĹؼ.˶⸴ԣео.һµķ˶.÷ض˶ʽܶԸӡ仯еλĴ˶и.˼ڽģ͵Ļϣ϶µ˳θ֣в˻ͼֵ˶Ԥͷֱͼƶ㷨.ʵ.
ؼʡͼУ˶ͼ֣ƶ
ͼšTP391.4
MODEL-BASED HUMAN MOTION TRACKING
LIU Xiao-Ming, ZHUANG Yue-Ting, and PAN Yun-He
(Institute of Artificial Intelligence, Zhejiang University, Hangzhou 310027)
AbstractIn the field of computer vision, the research on human motion analysis is attracting many researchers' attention with its wide application prospect. The key technical issue is the human motion tracking in the image sequence. Because of the complexity of human motion, the existing research methods impose limitation on human. A new approach to track the human motion is brought forward in this paper. Instead of aiming at a given human motion mode, it can track large motion from frame to frame in a complex, variational background. The main scheme is to track every part of human body from top to bottom on the basis of human model. The image-difference-based motion estimation and the divided region histogram similarity algorithm are used. The final result is given at the end of the paper.
Key wordsimage sequence, human motion, image difference, similarity 
1
ڼӾ˶ԽԽܵоߵ.һо൱㷺Ӧǰ˶Աҽѧȷǳ.⣬Զ⡢ʶ˶ܹ㷺Ӧڼӡ˻Ƶ.о˶زƵͼ.ˣͼʶ𲢸˶Ϊ˶Ĺؼ.ͳĸ˶ķһɣȴͼ֡ȡͲؽڵλã֮ʶȻڸ֡佨Ӧϵ.ģ֪ʶ֧֣ЩַΪ֣ģ֪ʶʱ15ݣֻҪͼģ֮佨ƥ䣬ôӦҲԶ.ûģ֪ʶʱ6ݣҪͨλáٶȡ״ȷӦ.˶⸴,ЩƣҪ󱳾һ˶ͶӰƽƽУ崩¿ȣ7.
һ·лģ͵˶.÷ȥеƣ粻ض˶ʽܶԸӡ仯еλĴ˶и.÷ûעһ֡ؽڵĻϣ˶Ԥαƥ䣬϶µ˳θ֣Ķά˶Ǽ.ù̵ͼ1ʾ.


ͼ1ϵͳ
ĵľ尲£2ڽܱõģͣ3ڽ˶ٵľ㷨4ڸʵĽ½ܽᲢ̽δĹ.
2ģ
ǽά忴ɹؽڵӵĸļϣ7.֫ؽӵ±壬ϱɼؽӵ.άǼ˶.ͶӰάƽϺ󣬹Ǽи߶εĿռϵȻֲ䣬Щ߶εĳȻᷢ仯ͶӰԭη˱仯.ڶάͼеУ߶βͼϢԱԶάģʾڶάƽϵͶӰͼ2ʾ.ε߼ΪͶӰĹǼܣ߽ӦĳηΪ.εĳΪͶӰĳȣѧ֪ʶȡ.


ͼ2άģ
ûԵһ֡ебעǾ͵õ˸֡иεͼϢ.ˣں֡ҵӦλãҲ͵õͶӰƽϵά˶Ǽ.û֡עи߿ɿԣʵϽȡӦ϶ΪһɣѰͼĹҲ˶ӦĽ.
3Ǽܸ
ͷڵ󣬿Էػȡɫ.ҵͷȷɳεһ㡪֪.ͬϱ۳ȷؽھ͹̶ˣ±ֻȷֵλü.ˣǴͷ㿪ʼ϶µ˳θ.水ͷɡ֫3ֱԵĸ.
3.1ͷ
ڶκһͼ֡ԣ֡˵ͷ˶ȡֲķʽôҵλÿܲȫŵ.ȡȫķʽôЧֻܵ.ˣǲȡַʽĽϣΪ˼һ֡ж˵ͷΧǲûͼֵ˶Ԥһ֡ͷλãȻԸԤΪģѡȡһ·αƥ䣬Ծȷλͷλ.
3.1.1ͼֵ˶Ԥ
ͼһӦù㷺ͼ8ݣڽʹԤ֡ͷ˶λ.赱ǰ֡kеͷ;֪ǽͷΪһԲHͼ3ʾHֱΪLԲΪO.һ֮֡ͷλƽСΪһOΪԲ,2LΪ뾶Բ.ʽɫֵ
DifferenceijR = |XijRed-XijRed|(i,j)ʦ
(1)
DifferenceijG = |XijGreen-XijGreen|(i,j)ʦ
(2)
DifferenceijB = |XijBlue-XijBlue|(i,j)ʦ
(3)


ͼ3ͷʾͼ
Xijʾk֡Ϊi,jصRGBɫֵ.Xijʾk+1֡Ϊi,jĵɫֵ.һ֡ͷ˶Աж˶ȻٹλƵĴС.ͼ4ǲʾͼHΪһ֡еͷλ.ļ֮֡䱳仯٣˲гͷ֮⣬ֵӽΪ.水㷨˶


ͼ4
 Ӧ0=90㿪ʼԦn+1=n+5,õ36Ƕֵ
 ÿnڵֱӳ򦸵ıཻ߽õ36Ϊ4L߶Ln;
 ʽÿ߶Lnеɫܺͣ

(4)
 ȡΪʹSumnLnӦĽǶȣ
={n|Sumn=MAX(Sum0Sum36)}.
(5)
㷨ĸ˶ͷλӳڲӦǸ÷ϵɫֵܺ.ȷ˶ҪͷλƵĴС.ȡӦL\-n߶Σ߶еRGBɫֵʽͼ5ʾ.ԿһԵ˫ֵߣͼ߷ֵӦͼ5H뱳Ĳ֣ұ߷ֵӦH뱳Ĳ.ʵϴͷͶӰƽϵλƣ˫ĿӦ.ĵķȽLn߶еɫֵaȻһٷֱp(50%)Ϊֵͼ5һֱY=a.p,ֱ˫ֵཻ߶,D1D2.ȡD1D2ȵľֵΪͷλ.ȷ˶㷨ֻǽһ߶ϣδȷ߶εһ˶.ʵֻD1D2Oľ룬D1OԶͷD1ڷ˶֮D2ڷ˶.ͨͼֵķǾͿԵõһ֡ͷĹλ.αƥõͷ׼ȷλúǻҪһ˶Ƶ׼ȷ.ֹλƹӰٷֱֵp֮Сp.Ӧصֵ˶Ƶľȵõ.ף9УǽӦKalman˲ķͷ˶Ԥ.Kalman˲н϶ľԣķЧʸߵŵ.ڻͼֵ˶Ԥ֡ͼϢû˶ʷϢĳЩŽϴӰԤЧ.


ͼ5˶ϵ˫ֵ
3.1.2αƥ
ͼֽһ֡ͷԤ꣬һԸõΪѡһ·ƥ䣬Ծȷλͷλ.Ϊ˱ͼƥ䣬ǽͷͼϵΪһΣmnĳmΪͷ뾱ľ룬nͨѧͷε֪ʶ.mnɫϢRGBֵΪͷɫģͱһ֡ƥ.ͼϵͷάͷͶӰƽϵͶӰͷάռ˶ͶӰ״仯.һ֡ͷֱΪ΢бͷˣ.ǵĿƥα֮.Ϊˣǲһֱַͼƶ㷨.
αA={(x,y),m,n,},ͼ6ʾ.(x,y)ΪһߵĽ㣬mΪĳnΪĿΪxļн.вտA={(x,y),m,n,}ͱȽϿA={(x,y),m,n,ȡ}·ߵƶ.


ͼ6
 AΪ4Сb1,b2,b3,b4,ֱ4СɫֱͼH1,H2,H3,H4;
 AΪ4Сb1,b2,b3,b4,ֱ4СɫֱͼH1,H2,H3,H4;
 㣺
(6)
WkʾռıأkǽHk׼ֱͼnΪֱͼάMin()ʾֱͼ󽻲ֱͼԽƣ󽻵ĽԽӽ1֮Խӽ0.ˣSʾα֮ƶȶSԽƶԽ.÷ķΪ˿еɫϵķֲҲɫλϵķֲ.
һͼ֡жԣǽһ֡ѸٵͷΪտAǰ֡дٵͷΪ A.һ֮֡ͷתСȡȡΪ-ȡܦȡܦ+, ΪСĽǶֵ.ͷͶӰСΪţm-mmm+m; n-(n/m)mnn+(n/m)m.·еÿһѡ㣨x,yǶ{(x,y),m,n,ȡ}γɼA䣬ֱһ֡ͷƶȣСƶȵAϢ.󣬾СƶȵA伴Ϊǰ֡ͷλ.ͷɫյ仯ǻԼȨķӦظɫģͣ10.
3.2
ǶɲֵĳΪӾ㵽ľ룬ΪҼ粿֮ľ.ΪЧضλɵλãǽزֺΪһ.ͷȣɿĳһǰͬʱŵģԶԳͿӦֱ妤m릤nȻ÷ֱͼƶ㷨ȷɿλ.ʱɿĿΪ֡еļ.ȷ󣬸Ҳȷ.β㣬Ǽһת.תΪһ֡90%֮֡ľҲΪһ֡μ90%.
3.3֫
֫˶ԣǷһ.֫һ֮֡˶϶࣬ӦȲһԤһ֡֫ܵλãȻپȷλ.ԴΪǽÿһ֡ϥؽڵβĽǶֵΪ˶켣Ϣ.ÿԤʱȡǰ֡ĽǶֵֵԸֵΪ֡ĳʼǶֵȡ䣬Ȼڣȡ-,ȡ+ȣķΧȷȷĽǶֵ.֣֫Ĵ˶ԤЧؼٿƥΧ.
֫αĲУ˦֮⣬һҪm.Ϊ֫ԣκӽͶӰȥȱ仯Сȿܱ仯ܶ.ÿθ֫ʱx,yѶn䣬mҪǰƶ㷨ȷ.
֫ĸУһ뿼ǵڵĴ.֫Ϊĳһ֡֫ڵڿƥзӳϵ͵ƶ.ڵƶֻ.˽ƥƶSͬʱΪƥĿŶȶ.֡еƥбÿ֫ƥŶȣϸƥŶȵ֮֡һ򼸸ϵͿŶȵ֡ø߿Ŷ֡йؽڵԲֵм伸֡Ĺؽڵ.ʵַ֤һ̶ϴڵŻ.
4ʵ
ݱ㷨, Ѿ΢ʵһԭͳ.ܶһͼе˶ֽи.һƵ40֡Ȼ䱣Ϊһͼ.ͼ7չʾ˸еĵ1510152025303540֡9ͼ.ͼϵ16ûʹעģ8֡ϵɳٵõ.ͼпԿͶӰϱ䳤£ÿ֡еͷɶܺõظٵ.ȻӵɫƣȲĸٻǱȽϾȷ.Ȼڱ㷨ûֵı߽ϢһЩؽڵ㣨40֡ףĸһƫ.ʵǻ֣ͼƽˮƽƶʱǵ㷨ҲܲܺõĸЧ.Ҫοʵʣhttp://icad.zju.edu.cn/liuxm/animation.html.


ͼ7ĸ
5
Ľһ·˶ͼе˶и.÷ȥе.ض˶ʽܶԸӡ仯еλĴ˶и.ĵķǳ㡢ֱ.ûֻڵһ֡ϱעؽڵ㣬ԺĹɼ.ǲȡ˴ͷ㿪ʼ϶µ˳θֵķ徭ͶӰĶά˶Ǽ.ڸٹУǲ˻ͼֵ˶Ԥͷֱͼƶ㷨.ʵٵЧǺõ.һ㷨ǽһڹǼлָά˶ģͣ9ݣʹά˶ʵ֤㷨Ч.
ʵǷ֣ڴڵʱٵЧܴܵӰ.һ;һڸٹмûķֽʽٵЧǷǳõģǽάģ͵֪ʶָά٣Ϳ޶ȵؼûĴ.Ҳǽһоķ.
ллͯһӱе.
õȻѧ.
߼飺С, , 19753,˶ʿо,ҪоΪý弼Ƶ.
ׯԽͦ, , 19656,ʿ,, ҪоΪýݿ⡢CAD.
ƺ, , 194611,,ʿʦ,Ժʿ,ҪоΪ˼άCADϵͳGISý弼.
ߵλ㽭ѧ˹оݡ310027
ο
1Rourke J O, Badler N I. Model-based image analysis of human motion using constraint propagation. IEEE Trans Pattern Anal Mach Intell, 1980, 2(6): 522536
2Hogg D. A program to see a walking person. Image Vision Computing, 1983, 5(20)
3Rohr K. Incremental recognition of pedestrians from image sequences. In: Proc IEEE Comput Soc Conf Comput Vision and Pattern Recogn, New York: IEEE Computer Society, 1993
4Chen Z, Lee H J. Knowledge-guided visual perception of 3D human gait from a single image sequence. IEEE Trans on Systems, Man, and Cybernetics, 1992, 22(2): 336342
5Bharatkumar A G, Daigle K E, Pandy M G et al. Low limb kinematics of human walking with the medial axis transformation. In: Proc of IEEE Computer Society Workshop on Motion of No-Rigid and Articulated Objects, Austin, TX: IEEE Computer Society, 1994. 7076
6Niyogi S A, Adelson E H. Analyzing and recognizing walking figures in XYT. In: Proc CVPR, Seattle, WA: IEEE Signal Processing Society, 467474
7Aggarwal J K, Cai Q. Human motion analysis: A review. In: Proc of the IEEE Nonrigid and Articulated Motion Workshop 1997, Piscataway, NJ: IEEE Computer Society 1997. 90102
8Jain R. Extraction of motion information from peripheral processes. IEEE Trans PAMI, 1981, (5)
9Zhuang Yueting, Liu Xiaoming, Pan Yunhe. Video motion capture using feature tracking and skeleton reconstruction. In: Proc of IEEE Int Conf on Image Processing. Kobe, Japan: IEEE Signal Processing Society, 1999. 2528
10Ҧѫ. ɫͼʵʱٷ. ѧ, 1998, 21(6): 527532
(Liu Mingbao, Yao Hongxun, Gao Wen. Real-time human face tracking in color images. Chinese Journal of Computers. 1998, 21(6): 527532) 
ԭյڣ1999-01-22
޸ĸյڣ1999-06-14.
