Deformable Object Tracking

We address the problem of visual tracking of arbitrary objects that undergo significant scale and appearance changes. The classical tracking methods rely on the bounding box surrounding the target object. Regardless of the tracking approach, the use of bounding box quite often introduces background information. This information propagates in time and its accumulation quite often results in drift and tracking failure. This is particularly the case with the particle filtering approach that is often used for visual tracking. However, it always uses a bounding box around the object to compute features of the particle samples. Since this causes the drift, we propose to use segmentation for sampling. Relying on segmentation and computing the colour and gradient orientation histograms from these segmented particle samples allows the tracker to easily adapt to the object's deformations, occlusions, orientation, scale and appearance changes. We propose two particle sampling strategies based on segmentation. In the rest, segmentation is done for every propagated particle sample, while in the second only the strongest particle sample is segmented. Depending on this decision there is obviously a trade-o between speed and performance.

This project comes from a collaboration with EADS Innovations Works, Germany.

Contact Person: Vasileios Belagiannis (TUM), Falk Schubert (EADS).

Currently, we do not offer our code but you can contact us (belagian@in.tum.de) for evaluating on your sequences.

News: Check out our new tracker (Click here)

Dataset

You can download the evaluation dataset with the annotations HERE .

Evaluation Sequences	No. of Frames	Image Resolution
01. Actions 2 [29]	2113	960X540
02. Entrance	196	720X576
03. Exit 1	186	720X576
04. Exit 2	172	720X576
07. Bridge	55	720X576
09. Pedestrian 2 [26]	352	720X576
08. Pedestrian 1 [27]	379	640X480
05. Skiing [7]	81	640X360
11. Mountain-bike [7]	228	640X360
12. Motocross 2 [7]	23	640X360
10. Cliff-dive 1 [7]	76	400X226
06. UAV [28]	716	368X288
13. Head	231	320X240

Results

We have compared our algorithm with Hough-Track [7], MIL [18] and TLD [19].

• table results soon.

Extra Video Results

• More results.

Related Work

V. Belagiannis, F. Schubert, N. Navab, S. Ilic Segmentation Based Particle Filtering for Real-Time 2D Object Tracking 12th European Conference on Computer Vision (ECCV), Firenze, Italy, October 07-13, 2012 (bib)

References

[7] Godec, M., Roth, P., Bischof, H.: Hough-based tracking of non-rigid objects. In: ICCV. (2011)

[12] Bibby, C., Reid, I.: Robust real-time visual tracking using pixel-wise posteriors. ECCV (2008)

[18] Babenko, B., Yang, M., Belongie, S.: Visual tracking with online multiple instance learning. In: CVPR. (2009)

[19] Kalal, Z., Matas, J., Mikolajczyk, K.: Pn learning: Bootstrapping binary classiers by structural constraints. In: CVPR. (2010)

[26] Pellegrini, S., Ess, A., Schindler, K., Van Gool, L.: You'll never walk alone: Modeling social behavior for multi-target tracking. In: ICCV. (2009)

[27] Leibe, B., Schindler, K., Van Gool, L.: Coupled detection and trajectory estimation for multi-object tracking. In: ICCV. (2007)

[28] Ollero, A., Lacroix, S., Merino, L., Gancet, J., Wiklund, J., Remuss, V., Perez, I., Gutierrez, L., Viegas, D., Benitez, M., et al.: Multiple eyes in the skies: architecture and perception issues in the comets unmanned air vehicles project. Robotics & Automation Magazine, IEEE (2005)

[29] Lockheed-Martin: Ucf lockheed-martin uav dataset. http://vision.eecs.ucf.edu/aerial/index.html (2009)