"Robust Human Tracking using Statistical Human Model of Appearance Variation" Kiyoshi Hashimoto (Keio Univ), Tomoyuki Kagaya(Keio Univ), Hirokatsu Kataoka (Keio Univ), Yoshimitsu Aoki (Keio Univ)
| ||||
In our proposed method, we create two statistical models to represent human-like body shape. The first is an Active Omega Model(AOM) to detect humans and track the contour of head-shoulder(Omega shape) accurately. The second is a Main-parts Link Model (MLM) to track each body parts robustly. And we track humans by applying these models. There are 4 steps for (i) background subtraction: MoG, (ii) human detection: AOM fitting using foreground, (iii) human tracking: 5 separated parts tracking using MLM, (iv) pose estimation: rough pose is estimated from results of separate parts tracking. We can understand human's "posture" and "simple action" as a result of this method. |
![]() |
References
| ||
Copyright Hirokatsu Kataoka |