Evaluation of visual tracking in extremely low frame rate wide area motion imagery

Visual tracking in wide area motion imagery (WAMI) is an important problem in security related tasks. The extremely low frame rate and the large camera motion in such videos, however, introduce challenging constraints that distinguish the task from traditional image tracking. In this study we evaluate the performance of several state-of-the-art visual trackers on the wide area surveillance videos. Specifically, we compare five visual trackers on sequences selected from the Columbus Large Image Format (CLIF) dataset for the vehicle tracking task. The experiments are conducted in two configurations: one with background registration to compensate for camera motion, and the other without. We evaluate the tracking performances both qualitatively and quantitatively. The experimental results show that (1) traditional visual trackers meet problems in the wide area surveillance videos and (2) background registration helps enhance the tracking performances, although there exists needed improvement for operational robust performance. We expect the visual tracking evaluation on low-frame rate wide area surveillance videos to motivate future research to addresses the related challenges and provide annotated images for benchmark purposes.