We study the problem of registration for scales of moving object on the video data with application to the gait analysis. A study on the human gait is important in the fields of the biometrics study and the sports/health managements for planning optimal trainings. Olshen et al. (1989) proposed the bootstrap estimation for confidence intervals of the functional data with application to the gait cycle data observed by the motion capture system. Kaziska and Srivastava (2007) proposed the method based on the data matching techniques using the dynamic time warping of human silhouettes.
The motion capture system can give the precise measurements of trajectories of moving objects, but it requires the laboratory environments and we cannot be used this system in the field study. On the other hand, the video camera is handy to observe the gait motion in the field study, but such data has many restrictions on analysis based on the filming conditions. In particular, the video data filmed from the frontal-view is difficult to analyze, because the subject getting closer to the camera, and observed data includes the scale-changing parameters. To cope with this, Okusa et al. (2011) proposed a registration for scales of moving object using the method of nonlinear least squares with application to the gait analysis assuming constant speed.
In this study, focusing on the human gait cycles, we consider the human gait modeling based on simple gait structure. We estimate the parameters of the human gait cycles using the method of nonlinear least squares (Okusa et al. 2011). We can also show that estimated parameters may be used for the human gait recognition. We apply a nearest-neighbor classifier, using the estimated parameters, to perform the human gait recognition, and present results from an experiment involving 55 subjects. As a result, our method shows 96.3% recognition rate, that has the better performance compared to other methods.
 Olshen, R., Biden, E., Wyatt, M., & Sutherland, D. (1989). Gait analysis and the bootstrap. The Annals of Statistics, 17, 4, 1419-1440.
 Kaziska, D. & Srivastava, A. (2007). Gait-Based Human Recognition by Classication of Cyclostationary Processes on Nonlinear Shape Manifolds. Journal of the American Statistical Association, 102, 480, 1114-1124.
 Okusa, K., Kamakuta, T., and Murakami, H. (2011). A Statistical Regis- tration of Scales of Moving Objects with Application to Walking Data (in Japanese). Bulletin of the Japanese Society of Computational Statistics, 23, 2, (to appear).
Keywords: Video Data Analysis; Gait Analysis; Gait Authentication
Biography: Kosuke Okusa currently works as a Research Assistant and Graduate Student with the Statistical Data Analysis Laboratory, Graduate School of Science and Engineering, Chuo University, Kourakuen Campus, in Tokyo. His area of interest is the Pattern Recognition of Human Avtivity, Human Motion Analysis and Authentication, Statistical Video Data Analysis.