Articulated Hand Tracking by ICA-based Hand Model and Multiple Cameras

Abstract

Vision has the great potential to give the computers the ability of collecting information. In this chapter we study the tracking and capturing of 3-D free hand motions by computers. The hand has no markers and no special devices and no special conditions are required. This chapter presents three techniques for the vision-based tracking. The first one is the ICAbased motion analysis. The second is articulated hand motion tracking by multiple cameras. The third is particle filtering with prediction. A human hand has many joints and its high dimensionality makes it difficult to model hand motions. To make things easier, it is important to represent a hand motion in a low dimensional space. Principal component analysis (PCA) has been proposed to reduce the dimensionality. However, the PCA basis vectors only represent global features, which are not optimal for representing intrinsic features. This chapter proposes an efficient representation of hand motions by independent component analysis (ICA). The ICA basis vectors represent local features, each of which corresponds to the motion of a particular finger. This representation is more efficient in modeling hand motions for tracking and recognizing hand-finger gestures in an image sequence. This chapter also proposes a new technique to simultaneously estimate the global hand pose and the finger articulation imaged by multiple cameras. Tracking a free hand motion against a cluttered background is a difficult task. The first reason is that hand fingers are selfoccluding and the second reason is the high dimensionality of the problem. In order to solve these difficulties, we propose using calibrated multiple cameras and at the same time improving search efficiency by predicted particle filtering. The effectiveness of our methods is demonstrated by tracking free hand motions in real image sequences. The method is easily expanded for tracking human body motions in 3D.