Upper Body Tracking

Body Tracking is a problem to track the position and orientation of human body from video sequence. It is difficult due to such problem as the high dimensionality of the state space, the self occlusion, the appearance changes, etc.

Figure 1 (a) Upper Body Tracking (b) Articulated upper body (c) Our DBN upper body Model

We propose a Dynamic Bayesian Network (DBN) model for upper body tracking (Fig 1(c) ). Our model aims at handling all kinds of physically feasible body motions, and we incorporate in the DBN various generic physical and anatomical constraints on the parts of the upper body. So, unlike most existing upper body model, our model can track any kind of body motion rather than only some typical movement.

We also explicitly model part occlusion in the DBN model, which allows automatically detect the occurrence of self-occlusion and to minimize the effect of measurement errors on the tracking accuracy due to occlusion.

Moreover, our method can handle 2D and 3D upper body tracking within the same framework.

2D upper body tracking

Fig.2 Tracking upper body from 2D sequence

(Note that our generic model can be generalized to different person and different motion.)

Demos: Dancing Drinking

3D upper body tracking (Using 2 web cameras)

Demo: Tracking upper body from 3D sequence