Real-time object tracking on mobile phones

For a mobile landmark recognition and Augmented Reality system, real-time object tracking is crucial for the user experience. In this paper, an effective natural feature (low dimensional Haar-like feature) based method is proposed for real-time object tracking on mobile phones. This method can cope with image variations and alleviate the drift problem. The main idea is using an online two template switching strategy for estimating the geometrical transformation. Experiments on NOKIA N900 phone show the good performance of the proposed method and support our claim that our tracking method is effective and robust.