Planar object tracking that aims at estimating the 2D transformation of the planar object in videos.

Planar object tracking (see the above Figure 1 for illustration) is a critical computer vision problem and has drawn increasing interest owing to its key roles in robotics, augmented reality, etc. Despite rapid progress, its further development, especially in the deep learning era, is largely hindered due to the lack of large-scale challenging benchmarks. Addressing this, we propose PlanarTrack, a large-scale challenging planar object tracking benchmark. Specifically, PlanarTrack consists of 1,000 sequences with more than 490K images. All these videos are collected in complex unconstrained scenarios from the wild, which makes PlanarTrack, compared to existing benchmarks, more challenging but realistic for real-world applications. To the best of our knowledge, PlanarTrack, to date, is the largest and most challenging dataset dedicated to planar tracking. To analyze the proposed PlanarTrack, we evaluate 10 planar trackers and conduct comprehensive comparisons and in-depth analysis. Our results, not surprisingly, demonstrate that current top-performing planar trackers degenerate significantly on the challenging PlanarTrack and more efforts are needed to improve planar tracking in future. Besides, we further derive a variant named PlanarTrack-BB for generic tracking from PlanarTrack. Our evaluation of 10 generic trackers shows that, surprisingly, PlanarTrack-BB is even more challenging than several popular generic tracking benchmarks and more attention should be paid to handle such planar objects, though they are rigid.

Highlights:

• Scale: PlanarTrack is to date the largest planar tracking benchmark with 1,000 videos from the wild
• Diversity: PlanarTrack includes 1,000 planar targets that are unique in each video for tracking
• High quality: Each video in PlanarTrack is manually labeled with careful inspection and refinment
• Challenge: All the video sequences in PlanarTrack are captured in unconstrained conditions
• PlanarTrack_BB: A new PlanarTrack_BB is derived from PlanarTrack for geneirc tracking

Comparison of PlanarTrack with other existing palanr tracking benchmarks.

Comparison of PlanarTrack with popular POT-210/280 on statistics.

Statisctis on challenging factors and comparison of training/test sets.

Overall performance of evaluated trackers on PlanarTrack using precision and success. Please refer to the paper for details of the evaluated trackers.

Evaluation on the two most common and the most two difficult challenging factors using precision.

Comparison of different trackers on PlanarTrack and POT-210.

Evaluation on PlanarTrack_BB and comparison with LaSOT and TrackingNet.

Qualitative video results of top-six trackers on PlanarTrack.

License and Contact: The dataset of our PlanarTrack is available for non-commercial research purposes only. If you have any question about the usage of PlanarTrack, please contact Libo Zhang or Heng Fan. If you have any questions about the dataset, please contact Xinran Liu at liuxinran@iscas.ac.cn.