Real-Time Tracking with On-line Feature Selection

The main idea is to formulate the tracking problem as a binary classification task and to achieve robustness by continuously updating the current classifier of the target object with respect to the current surrounding background. For this purpose we use an on-line AdaBoost feature selection algorithm [1] for tracking. The distinct advantage of the method is its capability of updating a model (classifier) during tracking. This allows on the one hand that a classifier can adapt to any object and on the other hand to handle appearance changes (e.g. out of plane rotations, illumination changes) quite naturally. Moreover, depending on the background the algorithm selects the most discriminating features for tracking resulting in stable tracking results. By using fast computable features (e.g. Haar wavelets, integral orientation histograms, local binary patterns) the algorithm runs in real-time (more than 20 fps using a standard 1.6 GHz PC with 512 MB RAM). The main innovation of the proposed tracking approach is an on-line AdaBoost algorithm [1] which allows efficient updating of a classifier and makes an on-line selection of tracking features feasible. The principle of the tracking approach with a classifier is depicted in Figure 1. Since we are