On-line Multi-View Forests for Tracking

Proceedings Symposium of the German Association for Pattern Recognition

A successful approach to tracking is to on-line learn discriminative
classiers for the target objects. Although these trackingby-
detection approaches are usually fast and accurate they easily drift in
case of putative and self-enforced wrong updates. Recent work has shown
that classier-based trackers can be signicantly stabilized by applying
semi-supervised learning methods instead of supervised ones. In this paper,
we propose a novel on-line multi-view learning algorithm based on
random forests. The main idea of our approach is to incorporate multiview
learning inside random forests and update each tree with individual
label estimates for the unlabeled data. Our method is fast, easy to implement,
benets from parallel computing architectures and inherently
exploits multiple views for learning from unlabeled data. In the tracking
experiments, we outperform the state-of-the-art methods based on
boosting and random forests.