Online Feature Selection Using Mutual Information for Real-Time Multi-view Object Tracking

Abstract

It has been shown that features can be selected adaptively for object tracking in changing environments [1]. We propose to use the variance of Mutual Information [2] for online feature selection to acquire reliable features for tracking by making use of the images of the tracked object in previous frames to refine our model so that the refined model after online feature selection becomes more robust. The ability of our method to pick up reliable features in real time is demonstrated with multi-view object tracking. In addition, the projective warping of 2D features is used to track 3D objects in non-frontal views in real time. Transformed 2D features can approximate relatively flat object structures such as the two eyes in a face. In this paper, approximations to the transformed features using weak perspective projection are derived. Since features in non-frontal views are computed on-the-fly by projective transforms under weak perspective projection, our framework requires only frontal-view training samples to track objects in multiple views.